2024-09-09 21:36:09 +00:00
# include "arg.h"
2023-03-25 18:26:40 +00:00
# include "common.h"
# include "llama.h"
2024-09-09 21:36:09 +00:00
# include <array>
# include <atomic>
2023-03-28 16:48:20 +00:00
# include <cmath>
2023-08-28 15:59:39 +00:00
# include <cstdio>
# include <cstring>
2023-04-16 10:13:00 +00:00
# include <ctime>
2024-09-09 21:36:09 +00:00
# include <fstream>
# include <mutex>
# include <random>
2023-07-22 12:21:24 +00:00
# include <sstream>
2023-08-25 16:05:02 +00:00
# include <thread>
2023-08-28 15:59:39 +00:00
# include <vector>
2023-03-28 16:48:20 +00:00
2023-06-16 18:23:53 +00:00
# if defined(_MSC_VER)
# pragma warning(disable: 4244 4267) // possible loss of data
# endif
2023-08-28 15:59:39 +00:00
struct results_perplexity {
std : : vector < llama_token > tokens ;
double ppl_value ;
std : : vector < float > logits ;
std : : vector < float > probs ;
} ;
struct results_log_softmax {
double log_softmax ;
float logit ;
float prob ;
} ;
2023-09-15 19:38:27 +00:00
static void write_logfile (
const llama_context * ctx , const gpt_params & params , const llama_model * model ,
const struct results_perplexity & results
) {
2023-08-28 15:59:39 +00:00
if ( params . logdir . empty ( ) ) {
return ;
}
if ( params . hellaswag ) {
fprintf ( stderr , " %s: warning: logging results is not implemented for HellaSwag. No files will be written. \n " , __func__ ) ;
return ;
}
2024-05-22 17:04:20 +00:00
const std : : string timestamp = string_get_sortable_timestamp ( ) ;
2023-08-28 15:59:39 +00:00
2024-05-22 17:04:20 +00:00
const bool success = fs_create_directory_with_parents ( params . logdir ) ;
2023-08-28 15:59:39 +00:00
if ( ! success ) {
fprintf ( stderr , " %s: warning: failed to create logdir %s, cannot write logfile \n " ,
__func__ , params . logdir . c_str ( ) ) ;
return ;
}
const std : : string logfile_path = params . logdir + timestamp + " .yml " ;
FILE * logfile = fopen ( logfile_path . c_str ( ) , " w " ) ;
if ( logfile = = NULL ) {
fprintf ( stderr , " %s: failed to open logfile %s \n " , __func__ , logfile_path . c_str ( ) ) ;
return ;
}
fprintf ( logfile , " binary: main \n " ) ;
char model_desc [ 128 ] ;
llama_model_desc ( model , model_desc , sizeof ( model_desc ) ) ;
2024-05-22 17:04:20 +00:00
yaml_dump_non_result_info ( logfile , params , ctx , timestamp , results . tokens , model_desc ) ;
2023-08-28 15:59:39 +00:00
fprintf ( logfile , " \n " ) ;
fprintf ( logfile , " ###################### \n " ) ;
fprintf ( logfile , " # Perplexity Results # \n " ) ;
fprintf ( logfile , " ###################### \n " ) ;
fprintf ( logfile , " \n " ) ;
2024-05-22 17:04:20 +00:00
yaml_dump_vector_float ( logfile , " logits " , results . logits ) ;
2023-08-28 15:59:39 +00:00
fprintf ( logfile , " ppl_value: %f \n " , results . ppl_value ) ;
2024-05-22 17:04:20 +00:00
yaml_dump_vector_float ( logfile , " probs " , results . probs ) ;
2023-08-28 15:59:39 +00:00
2024-09-07 12:16:19 +00:00
llama_perf_dump_yaml ( logfile , ctx ) ;
2023-08-28 15:59:39 +00:00
fclose ( logfile ) ;
}
2023-09-15 19:38:27 +00:00
static std : : vector < float > softmax ( const std : : vector < float > & logits ) {
2023-03-28 16:48:20 +00:00
std : : vector < float > probs ( logits . size ( ) ) ;
2023-03-25 18:26:40 +00:00
float max_logit = logits [ 0 ] ;
2023-09-28 16:04:36 +00:00
for ( float v : logits ) {
max_logit = std : : max ( max_logit , v ) ;
}
2023-03-25 18:26:40 +00:00
double sum_exp = 0.0 ;
for ( size_t i = 0 ; i < logits . size ( ) ; i + + ) {
// Subtract the maximum logit value from the current logit value for numerical stability
2023-03-28 16:48:20 +00:00
const float logit = logits [ i ] - max_logit ;
const float exp_logit = expf ( logit ) ;
2023-03-25 18:26:40 +00:00
sum_exp + = exp_logit ;
probs [ i ] = exp_logit ;
}
2023-09-28 16:04:36 +00:00
for ( size_t i = 0 ; i < probs . size ( ) ; i + + ) {
probs [ i ] / = sum_exp ;
}
2023-03-25 18:26:40 +00:00
return probs ;
}
2023-09-15 19:38:27 +00:00
static results_log_softmax log_softmax ( int n_vocab , const float * logits , int tok ) {
2023-08-25 16:05:02 +00:00
float max_logit = logits [ 0 ] ;
2023-09-28 16:04:36 +00:00
for ( int i = 1 ; i < n_vocab ; + + i ) {
max_logit = std : : max ( max_logit , logits [ i ] ) ;
}
2023-08-25 16:05:02 +00:00
double sum_exp = 0.0 ;
2023-09-28 16:04:36 +00:00
for ( int i = 0 ; i < n_vocab ; + + i ) {
sum_exp + = expf ( logits [ i ] - max_logit ) ;
}
2023-08-28 15:59:39 +00:00
return { logits [ tok ] - max_logit - log ( sum_exp ) , logits [ tok ] , expf ( logits [ tok ] - max_logit ) / ( float ) sum_exp } ;
2023-08-25 16:05:02 +00:00
}
2024-01-22 14:10:14 +00:00
static inline int nearest_int ( float fval ) {
//assert(fval <= 4194303.f);
float val = fval + 12582912.f ;
int i ; memcpy ( & i , & val , sizeof ( int ) ) ;
return ( i & 0x007fffff ) - 0x00400000 ;
}
static double log_softmax ( int n_vocab , const float * logits , uint16_t * log_prob , int tok ) {
float max_logit = logits [ 0 ] ;
float min_logit = logits [ 0 ] ;
for ( int i = 1 ; i < n_vocab ; + + i ) {
max_logit = std : : max ( max_logit , logits [ i ] ) ;
min_logit = std : : min ( min_logit , logits [ i ] ) ;
}
min_logit = std : : max ( min_logit , max_logit - 16 ) ;
double sum_exp = 0.0 ;
for ( int i = 0 ; i < n_vocab ; + + i ) {
sum_exp + = expf ( logits [ i ] - max_logit ) ;
}
const float log_sum_exp = log ( sum_exp ) ;
const float min_log_prob = min_logit - max_logit - log_sum_exp ;
const float scale = ( max_logit - min_logit ) / 65535.f ;
float * d = ( float * ) log_prob ;
d [ 0 ] = scale ;
d [ 1 ] = min_log_prob ;
log_prob + = 4 ;
if ( scale ) {
const float inv_scale = 1 / scale ;
for ( int i = 0 ; i < n_vocab ; + + i ) {
log_prob [ i ] = logits [ i ] > min_logit ? nearest_int ( inv_scale * ( logits [ i ] - min_logit ) ) : 0 ;
}
} else {
std : : memset ( log_prob , 0 , n_vocab * sizeof ( uint16_t ) ) ;
}
return max_logit + log_sum_exp - logits [ tok ] ;
}
2023-09-15 19:38:27 +00:00
static void process_logits (
int n_vocab , const float * logits , const int * tokens , int n_token , std : : vector < std : : thread > & workers ,
double & nll , double & nll2 , float * logit_history , float * prob_history
) {
2023-08-25 16:05:02 +00:00
std : : mutex mutex ;
int counter = 0 ;
2023-08-28 15:59:39 +00:00
auto compute = [ & mutex , & counter , & nll , & nll2 , logit_history , prob_history , n_vocab , logits , tokens , n_token ] ( ) {
2023-09-28 16:04:36 +00:00
double local_nll = 0 ;
double local_nll2 = 0 ;
2023-08-25 16:05:02 +00:00
while ( true ) {
std : : unique_lock < std : : mutex > lock ( mutex ) ;
int i = counter + + ;
if ( i > = n_token ) {
nll + = local_nll ; nll2 + = local_nll2 ;
break ;
}
lock . unlock ( ) ;
2023-08-28 15:59:39 +00:00
const results_log_softmax results = log_softmax ( n_vocab , logits + i * n_vocab , tokens [ i + 1 ] ) ;
const double v = - results . log_softmax ;
2023-08-25 16:05:02 +00:00
local_nll + = v ;
local_nll2 + = v * v ;
2023-08-28 15:59:39 +00:00
logit_history [ i ] = results . logit ;
prob_history [ i ] = results . prob ;
2023-08-25 16:05:02 +00:00
}
} ;
2023-09-28 16:04:36 +00:00
for ( auto & w : workers ) {
w = std : : thread ( compute ) ;
}
2023-08-25 16:05:02 +00:00
compute ( ) ;
2023-09-28 16:04:36 +00:00
for ( auto & w : workers ) {
w . join ( ) ;
}
2023-08-25 16:05:02 +00:00
}
2024-01-22 14:10:14 +00:00
static void process_logits ( std : : ostream & out , int n_vocab , const float * logits , const int * tokens , int n_token ,
std : : vector < std : : thread > & workers , std : : vector < uint16_t > & log_probs , double & nll , double & nll2 ) {
std : : mutex mutex ;
const int nv = 2 * ( ( n_vocab + 1 ) / 2 ) + 4 ;
int counter = 0 ;
auto compute = [ & mutex , & counter , & log_probs , & nll , & nll2 , n_vocab , logits , tokens , n_token , nv ] ( ) {
double local_nll = 0 ;
double local_nll2 = 0 ;
while ( true ) {
std : : unique_lock < std : : mutex > lock ( mutex ) ;
int i = counter + + ;
if ( i > = n_token ) {
nll + = local_nll ; nll2 + = local_nll2 ;
break ;
}
lock . unlock ( ) ;
const double v = log_softmax ( n_vocab , logits + i * n_vocab , log_probs . data ( ) + i * nv , tokens [ i + 1 ] ) ;
local_nll + = v ;
local_nll2 + = v * v ;
}
} ;
for ( auto & w : workers ) {
w = std : : thread ( compute ) ;
}
compute ( ) ;
for ( auto & w : workers ) {
w . join ( ) ;
}
out . write ( ( const char * ) log_probs . data ( ) , n_token * nv * sizeof ( uint16_t ) ) ;
}
struct kl_divergence_result {
2024-04-30 21:36:27 +00:00
double sum_nll = 0.0 ;
double sum_nll2 = 0.0 ;
double sum_nll_base = 0.0 ;
double sum_nll_base2 = 0.0 ;
double sum_nll_nll_base = 0.0 ;
double sum_kld = 0.0 ;
double sum_kld2 = 0.0 ;
double sum_p_diff = 0.0 ;
double sum_p_diff2 = 0.0 ;
double sum_p_diff4 = 0.0 ;
float max_p_diff = 0.0f ;
size_t n_same_top = 0.0 ;
size_t count = 0.0 ;
2024-01-22 14:10:14 +00:00
} ;
2024-04-30 21:36:27 +00:00
static std : : pair < double , float > log_softmax ( int n_vocab , const float * logits , const uint16_t * base_log_prob , int tok , kl_divergence_result & kld ) {
2024-01-22 14:10:14 +00:00
float max_logit = logits [ 0 ] ;
2024-01-23 13:17:20 +00:00
int imax = 0 ;
2024-01-22 14:10:14 +00:00
for ( int i = 1 ; i < n_vocab ; + + i ) {
2024-01-23 13:17:20 +00:00
if ( logits [ i ] > max_logit ) {
max_logit = logits [ i ] ;
imax = i ;
}
2024-01-22 14:10:14 +00:00
}
double sum_exp = 0.0 ;
for ( int i = 0 ; i < n_vocab ; + + i ) {
sum_exp + = expf ( logits [ i ] - max_logit ) ;
}
const float log_sum_exp = log ( sum_exp ) ;
const float * d = ( const float * ) base_log_prob ;
const float scale = d [ 0 ] ;
const float min_log_prob = d [ 1 ] ;
base_log_prob + = 4 ;
2024-04-30 21:36:27 +00:00
const float nll = max_logit + log_sum_exp - logits [ tok ] ;
2024-01-22 14:10:14 +00:00
kld . sum_nll + = nll ;
kld . sum_nll2 + = nll * nll ;
2024-04-30 21:36:27 +00:00
const float nll_base = - ( scale * base_log_prob [ tok ] + min_log_prob ) ;
kld . sum_nll_base + = nll_base ;
kld . sum_nll_base2 + = nll_base * nll_base ;
kld . sum_nll_nll_base + = nll * nll_base ;
2024-01-22 14:10:14 +00:00
max_logit + = log_sum_exp ;
double sum = 0 ;
2024-01-23 13:17:20 +00:00
int imax_base = - 1 ;
float p_log_base_max = 0 ;
2024-01-22 14:10:14 +00:00
for ( int i = 0 ; i < n_vocab ; + + i ) {
const float p_log_base = scale * base_log_prob [ i ] + min_log_prob ;
2024-01-23 13:17:20 +00:00
if ( i = = 0 | | p_log_base > p_log_base_max ) {
p_log_base_max = p_log_base ;
imax_base = i ;
}
2024-01-22 14:10:14 +00:00
if ( p_log_base > - 16.f ) {
const float p_base = expf ( p_log_base ) ;
sum + = p_base * ( p_log_base - logits [ i ] + max_logit ) ;
}
}
kld . sum_kld + = sum ;
kld . sum_kld2 + = sum * sum ;
+ + kld . count ;
2024-01-23 13:17:20 +00:00
if ( imax = = imax_base ) + + kld . n_same_top ;
2024-04-30 21:36:27 +00:00
const float p_base = expf ( - nll_base ) ;
const float p = expf ( - nll ) ;
const float p_diff = p - p_base ;
kld . sum_p_diff + = p_diff ;
const double p_diff2 = p_diff * p_diff ;
kld . sum_p_diff2 + = p_diff2 ;
kld . sum_p_diff4 + = p_diff2 * p_diff2 ;
kld . max_p_diff = std : : max ( kld . max_p_diff , std : : fabs ( p_diff ) ) ;
return std : : make_pair ( sum , p_diff ) ;
2024-01-22 14:10:14 +00:00
}
static void process_logits ( int n_vocab , const float * logits , const int * tokens , int n_token ,
2024-01-23 13:17:20 +00:00
std : : vector < std : : thread > & workers , const std : : vector < uint16_t > & base_log_probs , kl_divergence_result & kld ,
2024-04-30 21:36:27 +00:00
float * kld_values , float * p_diff_values ) {
2024-01-22 14:10:14 +00:00
std : : mutex mutex ;
const int nv = 2 * ( ( n_vocab + 1 ) / 2 ) + 4 ;
int counter = 0 ;
2024-04-30 21:36:27 +00:00
auto compute = [ & mutex , & counter , & base_log_probs , & kld , n_vocab , logits , tokens , n_token , nv , kld_values , p_diff_values ] ( ) {
2024-01-22 14:10:14 +00:00
kl_divergence_result local_kld ;
while ( true ) {
std : : unique_lock < std : : mutex > lock ( mutex ) ;
int i = counter + + ;
if ( i > = n_token ) {
2024-04-30 21:36:27 +00:00
kld . sum_nll + = local_kld . sum_nll ;
kld . sum_nll2 + = local_kld . sum_nll2 ;
kld . sum_nll_base + = local_kld . sum_nll_base ;
kld . sum_nll_base2 + = local_kld . sum_nll_base2 ;
kld . sum_nll_nll_base + = local_kld . sum_nll_nll_base ;
kld . sum_kld + = local_kld . sum_kld ;
kld . sum_kld2 + = local_kld . sum_kld2 ;
kld . sum_p_diff + = local_kld . sum_p_diff ;
kld . sum_p_diff2 + = local_kld . sum_p_diff2 ;
kld . sum_p_diff4 + = local_kld . sum_p_diff4 ;
kld . n_same_top + = local_kld . n_same_top ;
kld . max_p_diff = std : : max ( kld . max_p_diff , local_kld . max_p_diff ) ;
kld . count + = local_kld . count ;
2024-01-22 14:10:14 +00:00
break ;
}
lock . unlock ( ) ;
2024-04-30 21:36:27 +00:00
std : : pair < double , float > v = log_softmax ( n_vocab , logits + i * n_vocab , base_log_probs . data ( ) + i * nv , tokens [ i + 1 ] , local_kld ) ;
kld_values [ i ] = ( float ) v . first ;
p_diff_values [ i ] = v . second ;
2024-01-22 14:10:14 +00:00
}
} ;
for ( auto & w : workers ) {
w = std : : thread ( compute ) ;
}
compute ( ) ;
for ( auto & w : workers ) {
w . join ( ) ;
}
}
2023-09-15 19:38:27 +00:00
static results_perplexity perplexity_v2 ( llama_context * ctx , const gpt_params & params ) {
2024-02-18 20:39:30 +00:00
// Download: https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
2023-08-23 09:56:42 +00:00
// Run `./perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
// Output: `perplexity: 13.5106 [114/114]`
// BOS tokens will be added for each chunk before eval
2024-08-15 07:23:23 +00:00
const bool add_bos = llama_add_bos_token ( llama_get_model ( ctx ) ) ;
GGML_ASSERT ( ! llama_add_eos_token ( llama_get_model ( ctx ) ) ) ;
2023-08-23 20:08:04 +00:00
fprintf ( stderr , " %s: tokenizing the input .. \n " , __func__ ) ;
2024-04-09 17:44:08 +00:00
std : : vector < llama_token > tokens = : : llama_tokenize ( ctx , params . prompt , true ) ;
2023-08-29 20:55:45 +00:00
2023-09-28 19:42:38 +00:00
const int n_ctx = llama_n_ctx ( ctx ) ;
if ( int ( tokens . size ( ) ) < 2 * n_ctx ) {
fprintf ( stderr , " %s: you need at least %d tokens to evaluate perplexity with a context of %d \n " , __func__ , 2 * n_ctx ,
n_ctx ) ;
2023-08-29 20:55:45 +00:00
fprintf ( stderr , " %s: the data file you provided tokenizes to only %zu tokens \n " , __func__ , tokens . size ( ) ) ;
return { std : : move ( tokens ) , 0. , { } , { } } ;
}
2023-09-28 16:04:36 +00:00
std : : vector < float > logit_history ;
std : : vector < float > prob_history ;
2023-08-28 15:59:39 +00:00
logit_history . resize ( tokens . size ( ) ) ;
prob_history . resize ( tokens . size ( ) ) ;
if ( params . ppl_stride < = 0 ) {
fprintf ( stderr , " %s: stride is %d but must be greater than zero! \n " , __func__ , params . ppl_stride ) ;
return { tokens , - 1 , logit_history , prob_history } ;
}
2023-08-23 09:56:42 +00:00
2023-09-28 19:42:38 +00:00
const int calc_chunk = n_ctx ;
2023-08-23 09:56:42 +00:00
fprintf ( stderr , " %s: have %zu tokens. Calculation chunk = %d \n " , __func__ , tokens . size ( ) , calc_chunk ) ;
if ( int ( tokens . size ( ) ) < = calc_chunk ) {
fprintf ( stderr , " %s: there are only %zu tokens, this is not enough for a context size of %d and stride %d \n " , __func__ ,
2023-09-28 19:42:38 +00:00
tokens . size ( ) , n_ctx , params . ppl_stride ) ;
2023-08-28 15:59:39 +00:00
return { tokens , - 1 , logit_history , prob_history } ;
2023-08-23 09:56:42 +00:00
}
const int n_chunk_max = ( tokens . size ( ) - calc_chunk + params . ppl_stride - 1 ) / params . ppl_stride ;
const int n_chunk = params . n_chunks < 0 ? n_chunk_max : std : : min ( params . n_chunks , n_chunk_max ) ;
2023-09-28 19:42:38 +00:00
const int n_vocab = llama_n_vocab ( llama_get_model ( ctx ) ) ;
2023-08-23 09:56:42 +00:00
const int n_batch = params . n_batch ;
int count = 0 ;
double nll = 0.0 ;
fprintf ( stderr , " %s: calculating perplexity over %d chunks, batch_size=%d \n " , __func__ , n_chunk , n_batch ) ;
for ( int i = 0 ; i < n_chunk ; + + i ) {
const int start = i * params . ppl_stride ;
const int end = start + calc_chunk ;
const int num_batches = ( calc_chunk + n_batch - 1 ) / n_batch ;
//fprintf(stderr, "%s: evaluating %d...%d using %d batches\n", __func__, start, end, num_batches);
std : : vector < float > logits ;
const auto t_start = std : : chrono : : high_resolution_clock : : now ( ) ;
2023-09-28 16:04:36 +00:00
// clear the KV cache
2023-10-29 17:31:40 +00:00
llama_kv_cache_clear ( ctx ) ;
2023-09-28 16:04:36 +00:00
2023-08-23 09:56:42 +00:00
for ( int j = 0 ; j < num_batches ; + + j ) {
const int batch_start = start + j * n_batch ;
const int batch_size = std : : min ( end - batch_start , n_batch ) ;
//fprintf(stderr, " Batch %d: starts at %d, size is %d, n_past is %d\n",j,batch_start,batch_size,j * n_batch);
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: use llama_batch.logits instead of relying on logits_all == true
2023-09-28 19:42:38 +00:00
if ( llama_decode ( ctx , llama_batch_get_one ( tokens . data ( ) + batch_start , batch_size , j * n_batch , 0 ) ) ) {
2023-08-23 09:56:42 +00:00
//fprintf(stderr, "%s : failed to eval\n", __func__);
2023-08-28 15:59:39 +00:00
return { tokens , - 1 , logit_history , prob_history } ;
2023-08-23 09:56:42 +00:00
}
// save original token and restore it after eval
const auto token_org = tokens [ batch_start ] ;
// add BOS token for the first batch of each chunk
2023-08-23 20:08:04 +00:00
if ( add_bos & & j = = 0 ) {
2023-10-23 19:40:03 +00:00
tokens [ batch_start ] = llama_token_bos ( llama_get_model ( ctx ) ) ;
2023-08-23 09:56:42 +00:00
}
const auto batch_logits = llama_get_logits ( ctx ) ;
logits . insert ( logits . end ( ) , batch_logits , batch_logits + batch_size * n_vocab ) ;
if ( j = = 0 ) {
tokens [ batch_start ] = token_org ;
}
}
const auto t_end = std : : chrono : : high_resolution_clock : : now ( ) ;
if ( i = = 0 ) {
const float t_total = std : : chrono : : duration < float > ( t_end - t_start ) . count ( ) ;
fprintf ( stderr , " %s: %.2f seconds per pass - ETA " , __func__ , t_total ) ;
int total_seconds = ( int ) ( t_total * n_chunk ) ;
if ( total_seconds > = 60 * 60 ) {
fprintf ( stderr , " %d hours " , total_seconds / ( 60 * 60 ) ) ;
total_seconds = total_seconds % ( 60 * 60 ) ;
}
fprintf ( stderr , " %.2f minutes \n " , total_seconds / 60.0 ) ;
}
//fprintf(stderr, "%s: using tokens %d...%d\n",__func__,params.n_ctx - params.ppl_stride + start, params.n_ctx + start);
2023-09-28 19:42:38 +00:00
for ( int j = n_ctx - params . ppl_stride - 1 ; j < n_ctx - 1 ; + + j ) {
2023-08-23 09:56:42 +00:00
// Calculate probability of next token, given the previous ones.
const std : : vector < float > tok_logits (
logits . begin ( ) + ( j + 0 ) * n_vocab ,
logits . begin ( ) + ( j + 1 ) * n_vocab ) ;
const float prob = softmax ( tok_logits ) [ tokens [ start + j + 1 ] ] ;
2023-08-28 15:59:39 +00:00
logit_history [ start + j + 1 ] = tok_logits [ tokens [ start + j + 1 ] ] ;
prob_history [ start + j + 1 ] = prob ;
2023-08-23 09:56:42 +00:00
nll + = - std : : log ( prob ) ;
+ + count ;
}
// perplexity is e^(average negative log-likelihood)
if ( params . ppl_output_type = = 0 ) {
printf ( " [%d]%.4lf, " , i + 1 , std : : exp ( nll / count ) ) ;
} else {
printf ( " %8d %.4lf \n " , i * params . ppl_stride , std : : exp ( nll / count ) ) ;
}
fflush ( stdout ) ;
}
printf ( " \n " ) ;
2023-08-28 15:59:39 +00:00
return { tokens , std : : exp ( nll / count ) , logit_history , prob_history } ;
2023-08-23 09:56:42 +00:00
}
2024-03-09 18:55:54 +00:00
static results_perplexity perplexity ( llama_context * ctx , const gpt_params & params , const int32_t n_ctx ) {
2023-08-23 09:56:42 +00:00
if ( params . ppl_stride > 0 ) {
2023-08-28 15:59:39 +00:00
return perplexity_v2 ( ctx , params ) ;
2023-08-23 09:56:42 +00:00
}
2024-02-18 20:39:30 +00:00
// Download: https://huggingface.co/datasets/ggml-org/ci/resolve/main/wikitext-2-raw-v1.zip
2024-06-12 23:41:52 +00:00
// Run `./llama-perplexity -m models/7B/ggml-model-q4_0.bin -f wiki.test.raw`
2023-03-25 18:26:40 +00:00
// Output: `perplexity: 13.5106 [114/114]`
2023-05-08 14:41:54 +00:00
// BOS tokens will be added for each chunk before eval
2023-08-23 20:08:04 +00:00
2024-08-15 07:23:23 +00:00
const bool add_bos = llama_add_bos_token ( llama_get_model ( ctx ) ) ;
GGML_ASSERT ( ! llama_add_eos_token ( llama_get_model ( ctx ) ) ) ;
2023-08-23 20:08:04 +00:00
2024-01-22 14:10:14 +00:00
std : : ofstream logits_stream ;
if ( ! params . logits_file . empty ( ) ) {
2024-02-02 14:15:30 +00:00
logits_stream . open ( params . logits_file . c_str ( ) , std : : ios : : binary ) ;
2024-01-22 14:10:14 +00:00
if ( ! logits_stream . is_open ( ) ) {
fprintf ( stderr , " %s: failed to open %s for writing \n " , __func__ , params . logits_file . c_str ( ) ) ;
return { } ;
}
fprintf ( stderr , " %s: saving all logits to %s \n " , __func__ , params . logits_file . c_str ( ) ) ;
logits_stream . write ( " _logits_ " , 8 ) ;
2024-02-02 14:15:30 +00:00
logits_stream . write ( reinterpret_cast < const char * > ( & n_ctx ) , sizeof ( n_ctx ) ) ;
2024-01-22 14:10:14 +00:00
}
2023-08-27 13:50:33 +00:00
auto tim1 = std : : chrono : : high_resolution_clock : : now ( ) ;
2023-08-23 20:08:04 +00:00
fprintf ( stderr , " %s: tokenizing the input .. \n " , __func__ ) ;
2024-04-09 17:44:08 +00:00
std : : vector < llama_token > tokens = : : llama_tokenize ( ctx , params . prompt , true ) ;
2023-03-25 18:26:40 +00:00
2023-08-27 13:50:33 +00:00
auto tim2 = std : : chrono : : high_resolution_clock : : now ( ) ;
fprintf ( stderr , " %s: tokenization took %g ms \n " , __func__ , 1e-3 * std : : chrono : : duration_cast < std : : chrono : : microseconds > ( tim2 - tim1 ) . count ( ) ) ;
2023-09-28 19:42:38 +00:00
if ( int ( tokens . size ( ) ) < 2 * n_ctx ) {
fprintf ( stderr , " %s: you need at least %d tokens to evaluate perplexity with a context of %d \n " , __func__ , 2 * n_ctx ,
n_ctx ) ;
2023-08-29 20:55:45 +00:00
fprintf ( stderr , " %s: the data file you provided tokenizes to only %zu tokens \n " , __func__ , tokens . size ( ) ) ;
return { std : : move ( tokens ) , 0. , { } , { } } ;
}
2023-08-28 15:59:39 +00:00
std : : vector < float > logit_history ;
logit_history . resize ( tokens . size ( ) ) ;
std : : vector < float > prob_history ;
prob_history . resize ( tokens . size ( ) ) ;
2023-09-28 19:42:38 +00:00
const int n_chunk_max = tokens . size ( ) / n_ctx ;
2023-05-08 14:41:54 +00:00
2023-07-18 11:24:43 +00:00
const int n_chunk = params . n_chunks < 0 ? n_chunk_max : std : : min ( params . n_chunks , n_chunk_max ) ;
2023-09-28 19:42:38 +00:00
const int n_vocab = llama_n_vocab ( llama_get_model ( ctx ) ) ;
2023-05-08 14:41:54 +00:00
const int n_batch = params . n_batch ;
2023-03-25 18:26:40 +00:00
2023-07-18 11:24:43 +00:00
int count = 0 ;
2023-03-28 16:48:20 +00:00
double nll = 0.0 ;
2023-08-25 16:05:02 +00:00
double nll2 = 0.0 ;
2023-07-18 11:24:43 +00:00
2024-01-19 09:02:39 +00:00
const int num_batches = ( n_ctx + n_batch - 1 ) / n_batch ;
2024-03-09 18:55:54 +00:00
const int n_seq = std : : max ( 1 , n_batch / n_ctx ) ;
GGML_ASSERT ( n_batch < n_ctx | | n_batch % n_ctx = = 0 ) ;
GGML_ASSERT ( params . n_ctx = = n_seq * n_ctx ) ;
llama_batch batch = llama_batch_init ( std : : min ( n_batch , n_ctx * n_seq ) , 0 , 1 ) ;
2024-01-19 09:02:39 +00:00
std : : vector < float > logits ;
if ( num_batches > 1 ) {
logits . reserve ( ( size_t ) n_ctx * n_vocab ) ;
}
2024-03-09 18:55:54 +00:00
fprintf ( stderr , " %s: calculating perplexity over %d chunks, n_ctx=%d, batch_size=%d, n_seq=%d \n " , __func__ , n_chunk , n_ctx , n_batch , n_seq ) ;
2023-05-08 14:41:54 +00:00
2023-08-25 16:05:02 +00:00
std : : vector < std : : thread > workers ( std : : thread : : hardware_concurrency ( ) - 1 ) ;
2024-01-22 14:10:14 +00:00
std : : vector < uint16_t > log_probs ;
if ( ! params . logits_file . empty ( ) ) {
logits_stream . write ( ( const char * ) & n_vocab , sizeof ( n_vocab ) ) ;
logits_stream . write ( ( const char * ) & n_chunk , sizeof ( n_chunk ) ) ;
logits_stream . write ( ( const char * ) tokens . data ( ) , n_chunk * n_ctx * sizeof ( tokens [ 0 ] ) ) ;
const int nv = 2 * ( ( n_vocab + 1 ) / 2 ) + 4 ;
log_probs . resize ( n_ctx * nv ) ;
}
2024-03-09 18:55:54 +00:00
// We get the logits for all the tokens in the context window (params.n_ctx)
// from llama_eval above. Now, based on https://huggingface.co/docs/transformers/perplexity,
// calculate the perplexity over the last half of the window (so the model always has
// some context to predict the token).
//
// We rely on the fact that attention in the forward pass only looks at previous
// tokens here, so the logits returned for each token are an accurate representation
// of what the model would have predicted at that point.
//
// Example, we have a context window of 512, we will compute perplexity for each of the
// last 256 tokens. Then, we split the input up into context window size chunks to
// process the entire prompt.
const int first = n_ctx / 2 ;
for ( int i = 0 ; i < n_chunk ; i + = n_seq ) {
2023-09-28 19:42:38 +00:00
const int start = i * n_ctx ;
const int end = start + n_ctx ;
2023-03-25 18:26:40 +00:00
2024-03-09 18:55:54 +00:00
const int n_seq_batch = std : : min ( n_seq , n_chunk - i ) ;
2023-05-08 14:41:54 +00:00
const auto t_start = std : : chrono : : high_resolution_clock : : now ( ) ;
2023-09-28 16:04:36 +00:00
// clear the KV cache
2023-10-29 17:31:40 +00:00
llama_kv_cache_clear ( ctx ) ;
2023-09-28 16:04:36 +00:00
2023-04-13 21:50:42 +00:00
for ( int j = 0 ; j < num_batches ; + + j ) {
2023-05-08 14:41:54 +00:00
const int batch_start = start + j * n_batch ;
const int batch_size = std : : min ( end - batch_start , n_batch ) ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_outputs = 0 ;
2024-03-09 18:55:54 +00:00
batch . n_tokens = 0 ;
for ( int seq = 0 ; seq < n_seq_batch ; seq + + ) {
int seq_start = batch_start + seq * n_ctx ;
2023-05-08 14:41:54 +00:00
2024-03-09 18:55:54 +00:00
// save original token and restore it after eval
const auto token_org = tokens [ seq_start ] ;
// add BOS token for the first batch of each chunk
if ( add_bos & & j = = 0 ) {
tokens [ seq_start ] = llama_token_bos ( llama_get_model ( ctx ) ) ;
}
for ( int k = 0 ; k < batch_size ; + + k ) {
const int idx = seq * n_ctx + k ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
batch . token [ idx ] = tokens [ seq_start + k ] ;
batch . pos [ idx ] = j * n_batch + k ;
batch . n_seq_id [ idx ] = 1 ;
batch . seq_id [ idx ] [ 0 ] = seq ;
batch . logits [ idx ] = batch . pos [ idx ] > = first ? 1 : 0 ;
n_outputs + = batch . logits [ idx ] ! = 0 ;
2024-03-09 18:55:54 +00:00
}
batch . n_tokens + = batch_size ;
// restore the original token in case it was set to BOS
tokens [ seq_start ] = token_org ;
2023-05-08 14:41:54 +00:00
}
2024-03-09 18:55:54 +00:00
if ( llama_decode ( ctx , batch ) ) {
2023-04-13 21:50:42 +00:00
fprintf ( stderr , " %s : failed to eval \n " , __func__ ) ;
2023-08-28 15:59:39 +00:00
return { tokens , - 1 , logit_history , prob_history } ;
2023-04-13 21:50:42 +00:00
}
2023-05-08 14:41:54 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
if ( num_batches > 1 & & n_outputs > 0 ) {
2024-01-19 09:02:39 +00:00
const auto * batch_logits = llama_get_logits ( ctx ) ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
logits . insert ( logits . end ( ) , batch_logits , batch_logits + n_outputs * n_vocab ) ;
2024-01-19 09:02:39 +00:00
}
2023-03-25 18:26:40 +00:00
}
2023-05-08 14:41:54 +00:00
2023-03-25 18:26:40 +00:00
if ( i = = 0 ) {
2024-03-13 17:54:21 +00:00
llama_synchronize ( ctx ) ;
const auto t_end = std : : chrono : : high_resolution_clock : : now ( ) ;
2023-05-08 14:41:54 +00:00
const float t_total = std : : chrono : : duration < float > ( t_end - t_start ) . count ( ) ;
fprintf ( stderr , " %s: %.2f seconds per pass - ETA " , __func__ , t_total ) ;
2024-03-09 18:55:54 +00:00
int total_seconds = ( int ) ( t_total * n_chunk / n_seq ) ;
2023-04-21 12:57:57 +00:00
if ( total_seconds > = 60 * 60 ) {
2023-05-08 14:41:54 +00:00
fprintf ( stderr , " %d hours " , total_seconds / ( 60 * 60 ) ) ;
2023-04-21 12:57:57 +00:00
total_seconds = total_seconds % ( 60 * 60 ) ;
}
2023-08-18 09:48:55 +00:00
fprintf ( stderr , " %.2f minutes \n " , total_seconds / 60.0 ) ;
2023-03-25 18:26:40 +00:00
}
2023-05-08 14:41:54 +00:00
2024-03-09 18:55:54 +00:00
for ( int seq = 0 ; seq < n_seq_batch ; seq + + ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
const float * all_logits = num_batches > 1 ? logits . data ( ) : llama_get_logits_ith ( ctx , seq * n_ctx + first ) ;
2024-03-09 18:55:54 +00:00
llama_token * tokens_data = tokens . data ( ) + start + seq * n_ctx + first ;
if ( ! params . logits_file . empty ( ) ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
process_logits ( logits_stream , n_vocab , all_logits ,
2024-03-09 18:55:54 +00:00
tokens_data , n_ctx - 1 - first ,
workers , log_probs , nll , nll2 ) ;
} else {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
process_logits ( n_vocab , all_logits ,
2024-03-09 18:55:54 +00:00
tokens_data , n_ctx - 1 - first ,
workers , nll , nll2 ,
logit_history . data ( ) + start + seq * n_ctx + first ,
prob_history . data ( ) + start + seq * n_ctx + first ) ;
}
count + = n_ctx - first - 1 ;
// perplexity is e^(average negative log-likelihood)
if ( params . ppl_output_type = = 0 ) {
printf ( " [%d]%.4lf, " , i + seq + 1 , std : : exp ( nll / count ) ) ;
} else {
double av = nll / count ;
double av2 = nll2 / count - av * av ;
if ( av2 > 0 ) av2 = sqrt ( av2 / ( count - 1 ) ) ;
printf ( " %8d %.4lf %4lf %4lf \n " , i * n_ctx , std : : exp ( nll / count ) , av , av2 ) ;
}
2023-08-23 09:56:42 +00:00
}
2023-03-25 18:26:40 +00:00
fflush ( stdout ) ;
2024-01-19 09:02:39 +00:00
logits . clear ( ) ;
2023-03-25 18:26:40 +00:00
}
printf ( " \n " ) ;
2023-08-28 15:59:39 +00:00
2023-08-25 16:05:02 +00:00
nll2 / = count ;
nll / = count ;
2023-08-28 15:59:39 +00:00
const double ppl = exp ( nll ) ;
2023-08-25 16:05:02 +00:00
nll2 - = nll * nll ;
if ( nll2 > 0 ) {
nll2 = sqrt ( nll2 / ( count - 1 ) ) ;
printf ( " Final estimate: PPL = %.4lf +/- %.5lf \n " , ppl , nll2 * ppl ) ;
} else {
printf ( " Unexpected negative standard deviation of log(prob) \n " ) ;
}
2023-08-28 15:59:39 +00:00
2024-03-09 18:55:54 +00:00
llama_batch_free ( batch ) ;
2023-08-28 15:59:39 +00:00
return { tokens , ppl , logit_history , prob_history } ;
2023-03-25 18:26:40 +00:00
}
2024-01-19 08:45:06 +00:00
static bool decode_helper ( llama_context * ctx , llama_batch & batch , std : : vector < float > & batch_logits , int32_t n_batch , int32_t n_vocab ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int prev_outputs = 0 ;
2024-01-19 08:45:06 +00:00
for ( int32_t i = 0 ; i < ( int32_t ) batch . n_tokens ; i + = n_batch ) {
const int32_t n_tokens = std : : min ( n_batch , ( int32_t ) ( batch . n_tokens - i ) ) ;
llama_batch batch_view = {
n_tokens ,
batch . token + i ,
nullptr ,
batch . pos + i ,
batch . n_seq_id + i ,
batch . seq_id + i ,
batch . logits + i ,
0 , 0 , 0 , // unused
} ;
const int ret = llama_decode ( ctx , batch_view ) ;
if ( ret ! = 0 ) {
LOG_TEE ( " failed to decode the batch, n_batch = %d, ret = %d \n " , n_batch , ret ) ;
return false ;
2023-08-21 08:11:31 +00:00
}
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_outputs = 0 ;
for ( int i = 0 ; i < n_tokens ; + + i ) {
n_outputs + = batch_view . logits [ i ] ! = 0 ;
}
memcpy ( batch_logits . data ( ) + prev_outputs * n_vocab , llama_get_logits ( ctx ) , n_outputs * n_vocab * sizeof ( float ) ) ;
prev_outputs + = n_outputs ;
2023-08-21 08:11:31 +00:00
}
2024-01-19 08:45:06 +00:00
return true ;
2023-08-21 08:11:31 +00:00
}
2024-01-20 15:08:08 +00:00
# define K_TOKEN_CHUNK 4
2024-01-19 09:39:11 +00:00
static void compute_logprobs ( const float * batch_logits , int n_vocab , std : : vector < std : : thread > & workers ,
2024-01-18 17:18:21 +00:00
const std : : vector < std : : pair < size_t , llama_token > > & eval_pairs , std : : vector < float > & eval_results ) {
if ( eval_results . size ( ) ! = eval_pairs . size ( ) ) {
eval_results . resize ( eval_pairs . size ( ) ) ;
}
if ( eval_pairs . empty ( ) ) return ;
2024-01-20 15:08:08 +00:00
size_t max_threads = std : : min ( ( eval_pairs . size ( ) + K_TOKEN_CHUNK - 1 ) / K_TOKEN_CHUNK , workers . size ( ) ) ;
2024-01-18 17:18:21 +00:00
std : : atomic < int > counter ( 0 ) ;
auto compute = [ & counter , & eval_pairs , & eval_results , batch_logits , n_vocab ] ( ) {
2024-01-20 15:08:08 +00:00
float local_logprobs [ K_TOKEN_CHUNK ] ;
2024-01-18 17:18:21 +00:00
while ( true ) {
2024-01-20 15:08:08 +00:00
size_t first = counter . fetch_add ( K_TOKEN_CHUNK , std : : memory_order_relaxed ) ;
2024-01-18 17:18:21 +00:00
if ( first > = eval_results . size ( ) ) break ;
2024-01-20 15:08:08 +00:00
size_t last = std : : min ( first + K_TOKEN_CHUNK , eval_results . size ( ) ) ;
2024-01-18 17:18:21 +00:00
for ( size_t i = first ; i < last ; + + i ) {
auto logits = batch_logits + eval_pairs [ i ] . first * n_vocab ;
float max_logit = logits [ 0 ] ;
for ( int j = 1 ; j < n_vocab ; + + j ) {
max_logit = std : : max ( max_logit , logits [ j ] ) ;
}
float sum_p = 0.f ;
for ( int j = 0 ; j < n_vocab ; + + j ) {
sum_p + = expf ( logits [ j ] - max_logit ) ;
}
local_logprobs [ i - first ] = logits [ eval_pairs [ i ] . second ] - max_logit - std : : log ( sum_p ) ;
}
std : : memcpy ( eval_results . data ( ) + first , local_logprobs , ( last - first ) * sizeof ( float ) ) ;
}
} ;
for ( size_t it = 0 ; it < max_threads ; + + it ) {
workers [ it ] = std : : thread ( compute ) ;
}
for ( size_t it = 0 ; it < max_threads ; + + it ) {
workers [ it ] . join ( ) ;
}
}
2023-09-15 19:38:27 +00:00
static void hellaswag_score ( llama_context * ctx , const gpt_params & params ) {
2023-07-28 18:25:36 +00:00
// Calculates hellaswag score (acc_norm) from prompt
//
// Data extracted from the HellaSwag validation dataset (MIT license) https://github.com/rowanz/hellaswag/blob/master/data/hellaswag_val.jsonl
// All used data fields are preprocessed as in https://github.com/EleutherAI/lm-evaluation-harness/blob/df3da98c5405deafd519c2ddca52bb7c3fe36bef/lm_eval/tasks/hellaswag.py#L62-L68
//
// All 10042 tasks should be extracted to keep the results standardized like other implementations.
//
// Datafile layout:
// ['??'] denotes json fields
// 6 lines per task:
// ['activity_label'] + ": " +['ctx'] - The first part of the query, the context
// ['label'] - The index the best common sense ending aka gold ending
// ['endings'][0] - Endings added to the first part of the query
// ['endings'][1]
// ['endings'][2]
// ['endings'][3]
2023-07-22 12:21:24 +00:00
std : : vector < std : : string > prompt_lines ;
std : : istringstream strstream ( params . prompt ) ;
std : : string line ;
while ( std : : getline ( strstream , line , ' \n ' ) ) {
prompt_lines . push_back ( line ) ;
}
2024-01-18 13:33:01 +00:00
if ( prompt_lines . size ( ) % 6 ! = 0 ) {
2023-07-28 18:25:36 +00:00
fprintf ( stderr , " %s : number of lines in prompt not a multiple of 6. \n " , __func__ ) ;
return ;
}
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
size_t hs_task_count = prompt_lines . size ( ) / 6 ;
2023-08-04 10:07:21 +00:00
fprintf ( stderr , " %s : loaded %zu tasks from prompt. \n " , __func__ , hs_task_count ) ;
2023-07-22 12:21:24 +00:00
2023-09-28 19:42:38 +00:00
const bool is_spm = llama_vocab_type ( llama_get_model ( ctx ) ) = = LLAMA_VOCAB_TYPE_SPM ;
2023-08-26 13:48:53 +00:00
fprintf ( stderr , " ================================= is_spm = %d \n " , is_spm ) ;
2023-08-23 20:08:04 +00:00
2024-01-21 12:42:44 +00:00
// The tasks should be randomized so the score stabilizes quickly.
bool randomize_tasks = true ;
2023-07-28 18:25:36 +00:00
// Number of tasks to use when computing the score
2024-01-18 13:33:01 +00:00
if ( params . hellaswag_tasks < hs_task_count ) {
2023-07-28 18:25:36 +00:00
hs_task_count = params . hellaswag_tasks ;
}
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
// The random seed should not impact the final result if the computation is done over enough tasks, so kept hardcoded for now
std : : mt19937 rng ( 1 ) ;
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
// Dataholder for hellaswag tasks
struct hs_data_t {
std : : string context ;
size_t gold_ending_idx ;
std : : string ending [ 4 ] ;
size_t ending_logprob_count [ 4 ] ;
double ending_logprob [ 4 ] ;
2024-01-18 13:33:01 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits ; // starting index of logits in the llama_batch
2024-01-18 13:33:01 +00:00
size_t common_prefix ; // max number of initial tokens that are the same in all sentences
size_t required_tokens ; // needed number of tokens to evaluate all 4 endings
std : : vector < llama_token > seq_tokens [ 4 ] ;
2023-07-28 18:25:36 +00:00
} ;
2023-07-22 12:21:24 +00:00
2023-08-04 10:07:21 +00:00
fprintf ( stderr , " %s : selecting %zu %s tasks. \n " , __func__ , hs_task_count , ( randomize_tasks ? " randomized " : " the first " ) ) ;
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
// Select and read data from prompt lines
2024-01-18 13:33:01 +00:00
std : : vector < hs_data_t > hs_data ( hs_task_count ) ;
for ( size_t i = 0 ; i < hs_task_count ; i + + ) {
2023-07-28 18:25:36 +00:00
size_t idx = i ;
2023-07-22 12:21:24 +00:00
2024-01-18 13:33:01 +00:00
auto & hs_cur = hs_data [ i ] ;
2023-07-28 18:25:36 +00:00
// Select a random example of those left in the prompt
if ( randomize_tasks ) {
std : : uniform_int_distribution < size_t > dist ( 0 , prompt_lines . size ( ) / 6 - 1 ) ;
idx = dist ( rng ) ;
2023-07-22 12:21:24 +00:00
}
2024-01-18 13:33:01 +00:00
hs_cur . context = prompt_lines [ idx * 6 ] ;
hs_cur . gold_ending_idx = std : : stoi ( prompt_lines [ idx * 6 + 1 ] ) ;
for ( size_t j = 0 ; j < 4 ; j + + ) {
hs_cur . ending [ j ] = prompt_lines [ idx * 6 + 2 + j ] ;
2024-04-09 17:44:08 +00:00
hs_cur . seq_tokens [ j ] = : : llama_tokenize ( ctx , hs_cur . context + " " + hs_cur . ending [ j ] , true ) ;
2023-07-22 12:21:24 +00:00
}
2024-01-18 13:33:01 +00:00
// determine the common prefix of the endings
hs_cur . common_prefix = 0 ;
for ( size_t k = 0 ; k < hs_cur . seq_tokens [ 0 ] . size ( ) ; k + + ) {
if ( hs_cur . seq_tokens [ 0 ] [ k ] ! = hs_cur . seq_tokens [ 1 ] [ k ] | |
hs_cur . seq_tokens [ 0 ] [ k ] ! = hs_cur . seq_tokens [ 2 ] [ k ] | |
hs_cur . seq_tokens [ 0 ] [ k ] ! = hs_cur . seq_tokens [ 3 ] [ k ] ) {
break ;
}
hs_cur . common_prefix + + ;
}
hs_cur . required_tokens = hs_cur . common_prefix +
hs_cur . seq_tokens [ 0 ] . size ( ) - hs_cur . common_prefix +
hs_cur . seq_tokens [ 1 ] . size ( ) - hs_cur . common_prefix +
hs_cur . seq_tokens [ 2 ] . size ( ) - hs_cur . common_prefix +
hs_cur . seq_tokens [ 3 ] . size ( ) - hs_cur . common_prefix ;
2024-04-09 17:44:08 +00:00
//GGML_ASSERT(hs_cur.common_prefix >= ::llama_tokenize(ctx, hs_cur.context, true).size());
2024-01-18 13:33:01 +00:00
2023-07-28 18:25:36 +00:00
// Delete the selected random example from the prompt
if ( randomize_tasks ) {
prompt_lines . erase ( std : : next ( prompt_lines . begin ( ) , idx * 6 ) , std : : next ( prompt_lines . begin ( ) , idx * 6 + 6 ) ) ;
}
}
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
fprintf ( stderr , " %s : calculating hellaswag score over selected tasks. \n " , __func__ ) ;
2024-01-18 13:33:01 +00:00
2023-07-28 18:25:36 +00:00
printf ( " \n task \t acc_norm \n " ) ;
2023-07-22 12:21:24 +00:00
2023-07-28 18:25:36 +00:00
double acc = 0.0f ;
2024-01-18 13:33:01 +00:00
2023-09-28 19:42:38 +00:00
const int n_vocab = llama_n_vocab ( llama_get_model ( ctx ) ) ;
2024-01-18 13:33:01 +00:00
const int n_ctx = llama_n_ctx ( ctx ) ;
const int n_batch = params . n_batch ;
2023-07-28 18:25:36 +00:00
2024-01-19 08:45:06 +00:00
const int max_tasks_per_batch = 32 ;
2024-03-11 15:49:47 +00:00
const int max_seq = std : : min ( 4 * max_tasks_per_batch , ( int ) llama_n_seq_max ( ctx ) ) ;
2023-08-26 13:48:53 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
llama_batch batch = llama_batch_init ( n_ctx , 0 , 4 ) ;
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
std : : vector < float > tok_logits ( n_vocab ) ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: this could be made smaller; it's currently the worst-case size
2024-01-19 08:45:06 +00:00
std : : vector < float > batch_logits ( n_vocab * n_ctx ) ;
2024-01-18 13:33:01 +00:00
2024-01-18 17:18:21 +00:00
std : : vector < std : : pair < size_t , llama_token > > eval_pairs ;
std : : vector < float > eval_results ;
std : : vector < std : : thread > workers ( std : : thread : : hardware_concurrency ( ) ) ;
2024-01-18 13:33:01 +00:00
for ( size_t i0 = 0 ; i0 < hs_task_count ; i0 + + ) {
int n_cur = 0 ;
2023-09-28 16:04:36 +00:00
2024-01-18 13:33:01 +00:00
size_t i1 = i0 ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits = 0 ; // this tells us how many logits were needed before this point in the batch
2024-01-18 13:33:01 +00:00
llama_batch_clear ( batch ) ;
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
// batch as much tasks as possible into the available context
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// each task has 4 unique sequence ids - one for each ending
2024-01-18 13:33:01 +00:00
// the common prefix is shared among the 4 sequences to save tokens
// we extract logits only from the last common token and from all ending tokens of each sequence
while ( n_cur + ( int ) hs_data [ i1 ] . required_tokens < = n_ctx ) {
auto & hs_cur = hs_data [ i1 ] ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_logits = 0 ;
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
const int s0 = 4 * ( i1 - i0 ) ;
if ( s0 + 4 > max_seq ) {
break ;
}
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
for ( size_t i = 0 ; i < hs_cur . common_prefix ; + + i ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
llama_batch_add ( batch , hs_cur . seq_tokens [ 0 ] [ i ] , i , { s0 + 0 , s0 + 1 , s0 + 2 , s0 + 3 } , false ) ;
2024-01-18 13:33:01 +00:00
}
batch . logits [ batch . n_tokens - 1 ] = true ; // we need logits for the last token of the common prefix
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
n_logits + = 1 ;
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
for ( int s = 0 ; s < 4 ; + + s ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
const size_t seq_tokens_size = hs_cur . seq_tokens [ s ] . size ( ) ;
// TODO: don't evaluate the last token of each sequence
for ( size_t i = hs_cur . common_prefix ; i < seq_tokens_size ; + + i ) {
const bool needs_logits = i < seq_tokens_size - 1 ;
llama_batch_add ( batch , hs_cur . seq_tokens [ s ] [ i ] , i , { s0 + s } , needs_logits ) ;
n_logits + = needs_logits ;
2024-01-18 13:33:01 +00:00
}
}
2023-08-20 13:44:46 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
hs_cur . i_logits = i_logits ;
i_logits + = n_logits ;
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
n_cur + = hs_data [ i1 ] . required_tokens ;
if ( + + i1 = = hs_task_count ) {
break ;
}
2023-08-20 13:44:46 +00:00
}
2024-01-18 13:33:01 +00:00
if ( i0 = = i1 ) {
fprintf ( stderr , " %s : task %zu does not fit in the context window \n " , __func__ , i0 ) ;
return ;
}
2023-08-20 13:44:46 +00:00
2024-01-18 13:33:01 +00:00
llama_kv_cache_clear ( ctx ) ;
2023-07-28 18:25:36 +00:00
2024-01-18 13:33:01 +00:00
// decode all tasks [i0, i1)
2024-01-19 08:45:06 +00:00
if ( ! decode_helper ( ctx , batch , batch_logits , n_batch , n_vocab ) ) {
2024-01-18 13:33:01 +00:00
fprintf ( stderr , " %s: llama_decode() failed \n " , __func__ ) ;
return ;
}
2023-07-28 18:25:36 +00:00
2024-01-18 17:18:21 +00:00
// Compute log-probs in parallel
// First we collect all tasks
eval_pairs . clear ( ) ;
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & hs_cur = hs_data [ i ] ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t li = 1 ; // skip the last logit of the common prefix (computed separately below)
2024-01-18 17:18:21 +00:00
for ( int s = 0 ; s < 4 ; + + s ) {
for ( size_t j = hs_cur . common_prefix ; j < hs_cur . seq_tokens [ s ] . size ( ) - 1 ; j + + ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
eval_pairs . emplace_back ( hs_cur . i_logits + li + + , hs_cur . seq_tokens [ s ] [ j + 1 ] ) ;
2024-01-18 17:18:21 +00:00
}
}
}
// Then we do the actual calculation
2024-01-19 09:39:11 +00:00
compute_logprobs ( batch_logits . data ( ) , n_vocab , workers , eval_pairs , eval_results ) ;
2024-01-18 17:18:21 +00:00
size_t ir = 0 ;
2024-01-18 13:33:01 +00:00
// compute the logprobs for each ending of the decoded tasks
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & hs_cur = hs_data [ i ] ;
2023-07-22 12:21:24 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// get the logits of the last token of the common prefix
std : : memcpy ( tok_logits . data ( ) , batch_logits . data ( ) + n_vocab * hs_cur . i_logits , n_vocab * sizeof ( float ) ) ;
2023-07-28 18:25:36 +00:00
2024-01-18 13:33:01 +00:00
const auto first_probs = softmax ( tok_logits ) ;
2023-07-28 18:25:36 +00:00
2024-01-18 13:33:01 +00:00
for ( int s = 0 ; s < 4 ; + + s ) {
hs_cur . ending_logprob_count [ s ] = 1 ;
hs_cur . ending_logprob [ s ] = std : : log ( first_probs [ hs_cur . seq_tokens [ s ] [ hs_cur . common_prefix ] ] ) ;
for ( size_t j = hs_cur . common_prefix ; j < hs_cur . seq_tokens [ s ] . size ( ) - 1 ; j + + ) {
2024-01-18 17:18:21 +00:00
hs_cur . ending_logprob [ s ] + = eval_results [ ir + + ] ;
2024-01-18 13:33:01 +00:00
hs_cur . ending_logprob_count [ s ] + + ;
}
hs_cur . ending_logprob [ s ] / = hs_cur . ending_logprob_count [ s ] ;
}
2023-07-22 12:21:24 +00:00
2024-01-18 13:33:01 +00:00
// Find the ending with maximum logprob
size_t ending_logprob_max_idx = 0 ;
double ending_logprob_max_val = hs_cur . ending_logprob [ 0 ] ;
for ( size_t s = 1 ; s < 4 ; s + + ) {
if ( hs_cur . ending_logprob [ s ] > ending_logprob_max_val ) {
ending_logprob_max_idx = s ;
ending_logprob_max_val = hs_cur . ending_logprob [ s ] ;
}
2023-07-28 18:25:36 +00:00
}
2023-07-22 12:21:24 +00:00
2024-01-18 13:33:01 +00:00
//printf("max logprob ending idx %lu, gold ending idx %lu\n", ending_logprob_max_idx, hs_cur.gold_ending_idx);
2023-07-22 12:21:24 +00:00
2024-01-18 13:33:01 +00:00
// If the gold ending got the maximum logprobe add one accuracy point
if ( ending_logprob_max_idx = = hs_cur . gold_ending_idx ) {
acc + = 1.0 ;
}
// Print the accumulated accuracy mean x 100
printf ( " %zu \t %.8lf \n " , i + 1 , acc / double ( i + 1 ) * 100.0 ) ;
fflush ( stdout ) ;
2023-07-28 18:25:36 +00:00
}
2024-01-18 13:33:01 +00:00
i0 = i1 - 1 ;
2023-07-22 12:21:24 +00:00
}
2024-01-18 13:33:01 +00:00
llama_batch_free ( batch ) ;
2023-07-28 18:25:36 +00:00
2023-07-22 12:21:24 +00:00
printf ( " \n " ) ;
}
2024-01-18 11:46:27 +00:00
struct winogrande_entry {
std : : string first ;
std : : string second ;
std : : array < std : : string , 2 > choices ;
int answer ;
2024-01-19 08:45:06 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits ;
2024-01-19 08:45:06 +00:00
size_t common_prefix ;
size_t required_tokens ;
size_t n_base1 ; // number of tokens for context + choice 1
size_t n_base2 ; // number of tokens for context + choice 2
std : : vector < llama_token > seq_tokens [ 2 ] ;
2024-01-18 11:46:27 +00:00
} ;
2024-06-04 18:23:39 +00:00
static std : : vector < winogrande_entry > load_winogrande_from_csv ( const std : : string & prompt ) {
2024-01-18 11:46:27 +00:00
std : : vector < winogrande_entry > result ;
std : : istringstream in ( prompt ) ;
std : : string line ;
std : : array < int , 4 > comma_pos ;
while ( true ) {
std : : getline ( in , line ) ;
if ( in . fail ( ) | | in . eof ( ) ) break ;
int ipos = 0 ;
bool quote_open = false ;
for ( int i = 0 ; i < int ( line . size ( ) ) ; + + i ) {
if ( ! quote_open ) {
if ( line [ i ] = = ' , ' ) {
comma_pos [ ipos + + ] = i ;
if ( ipos = = 4 ) break ;
}
else if ( line [ i ] = = ' " ' ) {
quote_open = true ;
}
}
else {
if ( line [ i ] = = ' " ' ) {
quote_open = false ;
}
}
}
if ( ipos ! = 4 ) {
printf ( " %s: failed to find comma separators in <%s> \n " , __func__ , line . c_str ( ) ) ;
continue ;
}
auto sentence = line [ comma_pos [ 0 ] + 1 ] = = ' " ' ? line . substr ( comma_pos [ 0 ] + 2 , comma_pos [ 1 ] - comma_pos [ 0 ] - 3 )
: line . substr ( comma_pos [ 0 ] + 1 , comma_pos [ 1 ] - comma_pos [ 0 ] - 1 ) ;
auto choice1 = line . substr ( comma_pos [ 1 ] + 1 , comma_pos [ 2 ] - comma_pos [ 1 ] - 1 ) ;
auto choice2 = line . substr ( comma_pos [ 2 ] + 1 , comma_pos [ 3 ] - comma_pos [ 2 ] - 1 ) ;
auto answer = line . substr ( comma_pos [ 3 ] + 1 , line . size ( ) - comma_pos [ 3 ] - 1 ) ;
auto index = line . substr ( 0 , comma_pos [ 0 ] ) ;
int where = 0 ;
for ( ; where < int ( sentence . size ( ) ) ; + + where ) {
if ( sentence [ where ] = = ' _ ' ) break ;
}
if ( where = = int ( sentence . size ( ) ) ) {
printf ( " %s: no _ in <%s> \n " , __func__ , sentence . c_str ( ) ) ;
continue ;
}
std : : istringstream stream ( answer . c_str ( ) ) ;
int i_answer ; stream > > i_answer ;
if ( stream . fail ( ) | | i_answer < 1 | | i_answer > 2 ) {
printf ( " %s: failed to parse answer <%s> \n " , __func__ , answer . c_str ( ) ) ;
continue ;
}
result . emplace_back ( ) ;
auto & wg = result . back ( ) ;
wg . first = sentence . substr ( 0 , where ) ;
wg . second = sentence . substr ( where + 1 , sentence . size ( ) - where - 1 ) ;
wg . choices [ 0 ] = std : : move ( choice1 ) ;
wg . choices [ 1 ] = std : : move ( choice2 ) ;
wg . answer = i_answer ;
}
return result ;
}
/*
* Evaluates the Winogrande score .
* Uses a CSV containing task index , dentence , choice 1 , choice 2 , answer ( 1 or 2 )
* You can get one such dataset from e . g . https : //huggingface.co/datasets/ikawrakow/winogrande-eval-for-llama.cpp
* As an example , the 1 st row in the above dataset is
*
* 0 , Sarah was a much better surgeon than Maria so _ always got the easier cases . , Sarah , Maria , 2
*
*/
static void winogrande_score ( llama_context * ctx , const gpt_params & params ) {
constexpr int k_min_trailing_ctx = 3 ;
auto data = load_winogrande_from_csv ( params . prompt ) ;
if ( data . empty ( ) ) {
fprintf ( stderr , " %s: no tasks \n " , __func__ ) ;
return ;
}
fprintf ( stderr , " %s : loaded %zu tasks from prompt. \n " , __func__ , data . size ( ) ) ;
if ( params . winogrande_tasks > 0 & & params . winogrande_tasks < data . size ( ) ) {
fprintf ( stderr , " %s : selecting %zu random tasks \n " , __func__ , params . winogrande_tasks ) ;
std : : mt19937 rng ( 1 ) ;
std : : vector < int > aux ( data . size ( ) ) ;
for ( int i = 0 ; i < int ( data . size ( ) ) ; + + i ) {
aux [ i ] = i ;
}
float scale = 1 / ( 1.f + ( float ) rng . max ( ) ) ;
std : : vector < winogrande_entry > selected ;
2024-01-18 18:49:00 +00:00
selected . resize ( params . winogrande_tasks ) ;
2024-01-18 11:46:27 +00:00
for ( int i = 0 ; i < int ( params . winogrande_tasks ) ; + + i ) {
int j = int ( scale * rng ( ) * aux . size ( ) ) ;
selected [ i ] = std : : move ( data [ aux [ j ] ] ) ;
aux [ j ] = aux . back ( ) ;
aux . pop_back ( ) ;
}
data = std : : move ( selected ) ;
}
2024-01-19 08:45:06 +00:00
fprintf ( stderr , " %s : tokenizing selected tasks \n " , __func__ ) ;
for ( auto & task : data ) {
2024-04-09 17:44:08 +00:00
task . seq_tokens [ 0 ] = : : llama_tokenize ( ctx , task . first + task . choices [ 0 ] + task . second , true ) ;
task . seq_tokens [ 1 ] = : : llama_tokenize ( ctx , task . first + task . choices [ 1 ] + task . second , true ) ;
2024-01-19 08:45:06 +00:00
task . common_prefix = 0 ;
for ( size_t k = 0 ; k < task . seq_tokens [ 0 ] . size ( ) ; k + + ) {
if ( task . seq_tokens [ 0 ] [ k ] ! = task . seq_tokens [ 1 ] [ k ] ) {
break ;
}
task . common_prefix + + ;
}
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: the last token of each of the sequences don't need to be evaluated
2024-01-19 08:45:06 +00:00
task . required_tokens = task . common_prefix +
task . seq_tokens [ 0 ] . size ( ) - task . common_prefix +
task . seq_tokens [ 1 ] . size ( ) - task . common_prefix ;
2024-04-09 17:44:08 +00:00
task . n_base1 = : : llama_tokenize ( ctx , task . first + task . choices [ 0 ] , true ) . size ( ) ;
task . n_base2 = : : llama_tokenize ( ctx , task . first + task . choices [ 1 ] , true ) . size ( ) ;
2024-01-19 08:45:06 +00:00
}
2024-01-18 11:46:27 +00:00
fprintf ( stderr , " %s : calculating winogrande score over selected tasks. \n " , __func__ ) ;
const int n_vocab = llama_n_vocab ( llama_get_model ( ctx ) ) ;
2024-01-19 08:45:06 +00:00
const int n_ctx = llama_n_ctx ( ctx ) ;
const int n_batch = params . n_batch ;
const int max_tasks_per_batch = 128 ;
2024-03-11 15:49:47 +00:00
const int max_seq = std : : min ( 2 * max_tasks_per_batch , ( int ) llama_n_seq_max ( ctx ) ) ;
2024-01-19 08:45:06 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
llama_batch batch = llama_batch_init ( n_ctx , 0 , 2 ) ;
2024-01-18 11:46:27 +00:00
std : : vector < float > tok_logits ( n_vocab ) ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: this could be made smaller; it's currently the worst-case size
2024-01-19 08:45:06 +00:00
std : : vector < float > batch_logits ( n_vocab * n_ctx ) ;
2024-01-18 11:46:27 +00:00
2024-01-19 09:39:11 +00:00
std : : vector < std : : pair < size_t , llama_token > > eval_pairs ;
std : : vector < float > eval_results ;
std : : vector < std : : thread > workers ( std : : thread : : hardware_concurrency ( ) ) ;
2024-01-18 11:46:27 +00:00
int n_correct = 0 ;
int n_done = 0 ;
2024-01-19 08:45:06 +00:00
for ( size_t i0 = 0 ; i0 < data . size ( ) ; i0 + + ) {
int n_cur = 0 ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
size_t i1 = i0 ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits = 0 ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
llama_batch_clear ( batch ) ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
while ( n_cur + ( int ) data [ i1 ] . required_tokens < = n_ctx ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_logits = 0 ;
2024-01-19 08:45:06 +00:00
const int s0 = 2 * ( i1 - i0 ) ;
if ( s0 + 2 > max_seq ) {
break ;
}
for ( size_t i = 0 ; i < data [ i1 ] . common_prefix ; + + i ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
llama_batch_add ( batch , data [ i1 ] . seq_tokens [ 0 ] [ i ] , i , { s0 + 0 , s0 + 1 } , false ) ;
2024-01-19 08:45:06 +00:00
}
batch . logits [ batch . n_tokens - 1 ] = true ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
n_logits + = 1 ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
for ( int s = 0 ; s < 2 ; + + s ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: end before the last token, no need to predict past the end of the sequences
2024-01-19 08:45:06 +00:00
for ( size_t i = data [ i1 ] . common_prefix ; i < data [ i1 ] . seq_tokens [ s ] . size ( ) ; + + i ) {
llama_batch_add ( batch , data [ i1 ] . seq_tokens [ s ] [ i ] , i , { s0 + s } , true ) ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
n_logits + = 1 ;
2024-01-19 08:45:06 +00:00
}
}
2024-01-18 11:46:27 +00:00
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
data [ i1 ] . i_logits = i_logits ;
i_logits + = n_logits ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
n_cur + = data [ i1 ] . required_tokens ;
if ( + + i1 = = data . size ( ) ) {
break ;
}
}
if ( i0 = = i1 ) {
fprintf ( stderr , " %s : task %zu does not fit in the context window \n " , __func__ , i0 ) ;
return ;
}
2024-01-18 11:46:27 +00:00
llama_kv_cache_clear ( ctx ) ;
2024-01-19 08:45:06 +00:00
// decode all tasks [i0, i1)
if ( ! decode_helper ( ctx , batch , batch_logits , n_batch , n_vocab ) ) {
fprintf ( stderr , " %s: llama_decode() failed \n " , __func__ ) ;
2024-01-18 11:46:27 +00:00
return ;
}
2024-01-19 09:39:11 +00:00
eval_pairs . clear ( ) ;
2024-01-19 08:45:06 +00:00
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & task = data [ i ] ;
const bool skip_choice =
task . seq_tokens [ 0 ] . size ( ) - task . common_prefix > k_min_trailing_ctx & &
task . seq_tokens [ 1 ] . size ( ) - task . common_prefix > k_min_trailing_ctx ;
const auto & n_base1 = skip_choice ? task . n_base1 : task . common_prefix ;
const int last_1st = task . seq_tokens [ 0 ] . size ( ) - n_base1 > 1 ? 1 : 0 ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t li = n_base1 - task . common_prefix ;
2024-01-19 08:45:06 +00:00
for ( size_t j = n_base1 - 1 ; j < task . seq_tokens [ 0 ] . size ( ) - 1 - last_1st ; + + j ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
eval_pairs . emplace_back ( task . i_logits + li + + , task . seq_tokens [ 0 ] [ j + 1 ] ) ;
2024-01-18 11:46:27 +00:00
}
2024-01-19 08:45:06 +00:00
const auto & n_base2 = skip_choice ? task . n_base2 : task . common_prefix ;
const int last_2nd = task . seq_tokens [ 1 ] . size ( ) - n_base2 > 1 ? 1 : 0 ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// FIXME: this uses the wrong first logits when not skipping the choice word
li = task . seq_tokens [ 0 ] . size ( ) - task . common_prefix + n_base2 - task . common_prefix ;
2024-01-19 08:45:06 +00:00
for ( size_t j = n_base2 - 1 ; j < task . seq_tokens [ 1 ] . size ( ) - 1 - last_2nd ; + + j ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
eval_pairs . emplace_back ( task . i_logits + li + + , task . seq_tokens [ 1 ] [ j + 1 ] ) ;
2024-01-18 11:46:27 +00:00
}
2024-01-19 09:39:11 +00:00
}
compute_logprobs ( batch_logits . data ( ) , n_vocab , workers , eval_pairs , eval_results ) ;
size_t ir = 0 ;
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & task = data [ i ] ;
const bool skip_choice =
task . seq_tokens [ 0 ] . size ( ) - task . common_prefix > k_min_trailing_ctx & &
task . seq_tokens [ 1 ] . size ( ) - task . common_prefix > k_min_trailing_ctx ;
2024-01-18 11:46:27 +00:00
2024-01-19 09:39:11 +00:00
float score_1st = 0 ;
const auto & n_base1 = skip_choice ? task . n_base1 : task . common_prefix ;
const int last_1st = task . seq_tokens [ 0 ] . size ( ) - n_base1 > 1 ? 1 : 0 ;
for ( size_t j = n_base1 - 1 ; j < task . seq_tokens [ 0 ] . size ( ) - 1 - last_1st ; + + j ) {
score_1st + = eval_results [ ir + + ] ;
2024-01-19 08:45:06 +00:00
}
2024-01-19 09:39:11 +00:00
score_1st / = ( task . seq_tokens [ 0 ] . size ( ) - n_base1 - last_1st ) ;
2024-01-18 11:46:27 +00:00
2024-01-19 09:39:11 +00:00
float score_2nd = 0 ;
const auto & n_base2 = skip_choice ? task . n_base2 : task . common_prefix ;
const int last_2nd = task . seq_tokens [ 1 ] . size ( ) - n_base2 > 1 ? 1 : 0 ;
for ( size_t j = n_base2 - 1 ; j < task . seq_tokens [ 1 ] . size ( ) - 1 - last_2nd ; + + j ) {
score_2nd + = eval_results [ ir + + ] ;
2024-01-19 08:45:06 +00:00
}
2024-01-19 09:39:11 +00:00
score_2nd / = ( task . seq_tokens [ 1 ] . size ( ) - n_base2 - last_2nd ) ;
2024-01-18 11:46:27 +00:00
2024-01-19 08:45:06 +00:00
int result = score_1st > score_2nd ? 1 : 2 ;
if ( result = = task . answer ) {
+ + n_correct ;
}
+ + n_done ;
2024-01-18 11:46:27 +00:00
2024-01-19 09:39:11 +00:00
// print the accumulated accuracy mean x 100
2024-01-19 08:45:06 +00:00
printf ( " %zu \t %.4lf \t %10.6f %10.6f %d %d \n " , i + 1 , 100.0 * n_correct / n_done , score_1st , score_2nd , result , task . answer ) ;
fflush ( stdout ) ;
2024-01-18 11:46:27 +00:00
}
2024-01-19 08:45:06 +00:00
i0 = i1 - 1 ;
2024-01-18 11:46:27 +00:00
}
printf ( " \n " ) ;
if ( n_done < 100 ) return ;
const float p = 1.f * n_correct / n_done ;
const float sigma = 100.f * sqrt ( p * ( 1 - p ) / ( n_done - 1 ) ) ;
printf ( " Final Winogrande score(%d tasks): %.4lf +/- %.4lf \n " , n_done , 100 * p , sigma ) ;
}
2024-01-23 12:12:57 +00:00
static bool deserialize_string ( std : : istream & in , std : : string & str ) {
2024-01-21 12:42:44 +00:00
uint32_t size ;
if ( ! in . read ( ( char * ) & size , sizeof ( size ) ) . fail ( ) ) {
str . resize ( size ) ;
2024-01-23 12:12:57 +00:00
if ( ! in . read ( ( char * ) & str [ 0 ] , size ) . fail ( ) ) return true ;
2024-01-21 12:42:44 +00:00
}
return false ;
}
struct multiple_choice_answers {
std : : vector < std : : string > answers ;
std : : vector < int > labels ;
bool deserialize ( std : : istream & in ) {
uint32_t n ;
in . read ( ( char * ) & n , sizeof ( n ) ) ;
if ( in . fail ( ) | | n > 100 ) return false ; // 100 as max. number of answers should be good enough for any practical purpose
answers . resize ( n ) ;
labels . resize ( n ) ;
for ( auto & a : answers ) {
if ( ! deserialize_string ( in , a ) ) return false ;
}
in . read ( ( char * ) labels . data ( ) , n * sizeof ( int ) ) ;
return ! in . fail ( ) ;
}
} ;
struct multiple_choice_task {
std : : string question ; // the question (or context that needs to be continued)
multiple_choice_answers mc1 ; // possible answers (continuations) with a single correct answer
multiple_choice_answers mc2 ; // possible answers (continuations) with multiple correct answers - not handled yet
bool deserialize ( std : : istream & in ) {
if ( ! deserialize_string ( in , question ) ) return false ;
return mc1 . deserialize ( in ) & & mc2 . deserialize ( in ) ;
}
// For evaluation
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits ; // starting index of logits in the llama_batch
2024-01-21 12:42:44 +00:00
size_t common_prefix ; // max number of initial tokens that are the same in all sentences
size_t required_tokens ; // needed number of tokens to evaluate all answers
std : : vector < std : : vector < llama_token > > seq_tokens ;
std : : vector < float > log_probs ;
} ;
2024-04-09 17:44:08 +00:00
static bool multiple_choice_prepare_one_task ( llama_context * ctx , multiple_choice_task & task , bool log_error ) {
2024-01-21 12:42:44 +00:00
if ( task . question . empty ( ) | | task . mc1 . answers . empty ( ) ) {
if ( log_error ) {
printf ( " %s: found bad task with empty question and/or answers \n " , __func__ ) ;
}
return false ;
}
task . seq_tokens . reserve ( task . mc1 . answers . size ( ) ) ;
for ( auto & answer : task . mc1 . answers ) {
if ( answer . empty ( ) ) {
if ( log_error ) {
printf ( " %s: found empty answer \n " , __func__ ) ;
}
return false ;
}
2024-04-09 17:44:08 +00:00
task . seq_tokens . emplace_back ( : : llama_tokenize ( ctx , task . question + " " + answer , true ) ) ;
2024-01-21 12:42:44 +00:00
}
auto min_len = task . seq_tokens . front ( ) . size ( ) ;
for ( auto & seq : task . seq_tokens ) {
min_len = std : : min ( min_len , seq . size ( ) ) ;
}
task . common_prefix = 0 ;
for ( size_t k = 0 ; k < min_len ; + + k ) {
auto token = task . seq_tokens [ 0 ] [ k ] ;
bool all_same = true ;
for ( size_t i = 1 ; i < task . seq_tokens . size ( ) ; + + i ) {
if ( task . seq_tokens [ i ] [ k ] ! = token ) {
all_same = false ;
break ;
}
}
if ( ! all_same ) {
break ;
}
+ + task . common_prefix ;
}
task . required_tokens = task . common_prefix ;
for ( auto & seq : task . seq_tokens ) {
task . required_tokens + = seq . size ( ) - task . common_prefix ;
}
return true ;
}
//
// Calculates score for multiple choice tasks with single correct answer from prompt.
// Commonly used LLM evaluation metrics of this type are
// * ARC
// * HellaSwag
// * MMLU
// * TruthfulQA
//
// Validation datasets for these 4 tests can be found at
// https://huggingface.co/datasets/ikawrakow/validation-datasets-for-llama.cpp
// The data for these datasets was extracted from
// git@hf.co:datasets/allenai/ai2_arc
// https://github.com/rowanz/hellaswag/blob/master/data/hellaswag_val.jsonl
// git@hf.co:datasets/Stevross/mmlu
// https://huggingface.co/datasets/truthful_qa
//
static void multiple_choice_score ( llama_context * ctx , const gpt_params & params ) {
std : : istringstream strstream ( params . prompt ) ;
uint32_t n_task ;
strstream . read ( ( char * ) & n_task , sizeof ( n_task ) ) ;
if ( strstream . fail ( ) | | n_task = = 0 ) {
printf ( " %s: no tasks \n " , __func__ ) ;
return ;
}
printf ( " %s: there are %u tasks in prompt \n " , __func__ , n_task ) ;
std : : vector < uint32_t > task_pos ( n_task ) ;
strstream . read ( ( char * ) task_pos . data ( ) , task_pos . size ( ) * sizeof ( uint32_t ) ) ;
if ( strstream . fail ( ) ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
printf ( " %s: failed to read task positions from prompt \n " , __func__ ) ;
2024-01-21 12:42:44 +00:00
return ;
}
std : : vector < multiple_choice_task > tasks ;
if ( params . multiple_choice_tasks = = 0 | | params . multiple_choice_tasks > = ( size_t ) n_task ) {
// Use all tasks
tasks . resize ( n_task ) ;
printf ( " %s: reading tasks " , __func__ ) ;
2024-05-18 07:57:08 +00:00
int n_dot = std : : max ( ( int ) n_task / 100 , 1 ) ;
2024-01-21 12:42:44 +00:00
int i = 0 ;
for ( auto & task : tasks ) {
+ + i ;
if ( ! task . deserialize ( strstream ) ) {
printf ( " %s: failed to read task %d of %u \n " , __func__ , i , n_task ) ;
return ;
}
if ( i % n_dot = = 0 ) printf ( " . " ) ;
}
printf ( " done \n " ) ;
}
else {
printf ( " %s: selecting %zu random tasks from %u tasks available \n " , __func__ , params . multiple_choice_tasks , n_task ) ;
std : : mt19937 rng ( 1 ) ;
std : : vector < int > aux ( n_task ) ;
for ( uint32_t i = 0 ; i < n_task ; + + i ) aux [ i ] = i ;
float scale = 1.f / ( 1.f + ( float ) std : : mt19937 : : max ( ) ) ;
tasks . resize ( params . multiple_choice_tasks ) ;
for ( auto & task : tasks ) {
int j = ( int ) ( scale * rng ( ) * aux . size ( ) ) ;
int idx = aux [ j ] ;
aux [ j ] = aux . back ( ) ;
aux . pop_back ( ) ;
strstream . seekg ( task_pos [ idx ] , std : : ios : : beg ) ;
if ( ! task . deserialize ( strstream ) ) {
printf ( " %s: failed to read task %d at position %u \n " , __func__ , idx , task_pos [ idx ] ) ;
return ;
}
}
n_task = params . multiple_choice_tasks ;
}
printf ( " %s: preparing task data " , __func__ ) ;
fflush ( stdout ) ;
if ( n_task > 500 ) {
printf ( " ... " ) ;
fflush ( stdout ) ;
std : : atomic < int > counter ( 0 ) ;
std : : atomic < int > n_bad ( 0 ) ;
2024-04-09 17:44:08 +00:00
auto prepare = [ & counter , & n_bad , & tasks , ctx ] ( ) {
2024-01-21 12:42:44 +00:00
int num_tasks = tasks . size ( ) ;
int n_bad_local = 0 ;
while ( true ) {
int first = counter . fetch_add ( K_TOKEN_CHUNK ) ;
if ( first > = num_tasks ) {
if ( n_bad_local > 0 ) n_bad + = n_bad_local ;
break ;
}
int last = std : : min ( first + K_TOKEN_CHUNK , num_tasks ) ;
for ( int i = first ; i < last ; + + i ) {
2024-04-09 17:44:08 +00:00
if ( ! multiple_choice_prepare_one_task ( ctx , tasks [ i ] , false ) ) + + n_bad_local ;
2024-01-21 12:42:44 +00:00
}
}
} ;
size_t max_thread = std : : thread : : hardware_concurrency ( ) ;
max_thread = std : : min ( max_thread , ( tasks . size ( ) + K_TOKEN_CHUNK - 1 ) / K_TOKEN_CHUNK ) ;
std : : vector < std : : thread > workers ( max_thread - 1 ) ;
for ( auto & w : workers ) w = std : : thread ( prepare ) ;
prepare ( ) ;
for ( auto & w : workers ) w . join ( ) ;
printf ( " done \n " ) ;
fflush ( stdout ) ;
int nbad = n_bad ;
if ( nbad > 0 ) {
printf ( " %s: found %d malformed tasks \n " , __func__ , nbad ) ;
return ;
}
} else {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_dot = std : : max ( ( int ) n_task / 100 , 1 ) ;
2024-01-21 12:42:44 +00:00
int i_task = 0 ;
for ( auto & task : tasks ) {
+ + i_task ;
2024-04-09 17:44:08 +00:00
if ( ! multiple_choice_prepare_one_task ( ctx , task , true ) ) {
2024-01-21 12:42:44 +00:00
return ;
}
if ( i_task % n_dot = = 0 ) {
printf ( " . " ) ;
fflush ( stdout ) ;
}
}
printf ( " done \n " ) ;
}
printf ( " %s : calculating TruthfulQA score over %zu tasks. \n " , __func__ , tasks . size ( ) ) ;
printf ( " \n task \t acc_norm \n " ) ;
const int n_vocab = llama_n_vocab ( llama_get_model ( ctx ) ) ;
const int n_ctx = llama_n_ctx ( ctx ) ;
const int n_batch = params . n_batch ;
const int max_tasks_per_batch = 32 ;
2024-03-11 15:49:47 +00:00
const int max_seq = std : : min ( 4 * max_tasks_per_batch , ( int ) llama_n_seq_max ( ctx ) ) ;
2024-01-21 12:42:44 +00:00
llama_batch batch = llama_batch_init ( n_ctx , 0 , max_seq ) ;
std : : vector < float > tok_logits ( n_vocab ) ;
std : : vector < float > batch_logits ( n_vocab * n_ctx ) ;
std : : vector < std : : pair < size_t , llama_token > > eval_pairs ;
std : : vector < float > eval_results ;
std : : vector < std : : thread > workers ( std : : thread : : hardware_concurrency ( ) ) ;
std : : vector < int > batch_indeces ;
int n_done = 0 ;
int n_correct = 0 ;
int n_tot_answers = 0 ;
for ( size_t i0 = 0 ; i0 < tasks . size ( ) ; i0 + + ) {
int n_cur = 0 ;
size_t i1 = i0 ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t i_logits = 0 ; // this tells us how many logits were needed before this point in the batch
2024-01-21 12:42:44 +00:00
llama_batch_clear ( batch ) ;
// batch as much tasks as possible into the available context
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// each task has 4 unique sequence ids - one for each ending
2024-01-21 12:42:44 +00:00
// the common prefix is shared among the 4 sequences to save tokens
// we extract logits only from the last common token and from all ending tokens of each sequence
int s0 = 0 ;
while ( n_cur + ( int ) tasks [ i1 ] . required_tokens < = n_ctx ) {
auto & cur_task = tasks [ i1 ] ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
int n_logits = 0 ;
2024-01-21 12:42:44 +00:00
int num_answers = cur_task . seq_tokens . size ( ) ;
if ( s0 + num_answers > max_seq ) {
break ;
}
if ( int ( batch_indeces . size ( ) ) ! = num_answers ) {
batch_indeces . resize ( num_answers ) ;
}
for ( int s = 0 ; s < num_answers ; + + s ) batch_indeces [ s ] = s0 + s ;
for ( size_t i = 0 ; i < cur_task . common_prefix ; + + i ) {
//llama_batch_add(batch, cur_task.seq_tokens[0][i], i, { s0 + 0, s0 + 1, s0 + 2, s0 + 3}, false);
llama_batch_add ( batch , cur_task . seq_tokens [ 0 ] [ i ] , i , batch_indeces , false ) ;
}
batch . logits [ batch . n_tokens - 1 ] = true ; // we need logits for the last token of the common prefix
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
n_logits + = 1 ;
2024-01-21 12:42:44 +00:00
for ( int s = 0 ; s < int ( cur_task . seq_tokens . size ( ) ) ; + + s ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
const size_t seq_tokens_size = cur_task . seq_tokens [ s ] . size ( ) ;
// TODO: don't evaluate the last token of each sequence
for ( size_t i = cur_task . common_prefix ; i < seq_tokens_size ; + + i ) {
const bool needs_logits = i < seq_tokens_size - 1 ;
llama_batch_add ( batch , cur_task . seq_tokens [ s ] [ i ] , i , { s0 + s } , needs_logits ) ;
n_logits + = needs_logits ;
2024-01-21 12:42:44 +00:00
}
}
s0 + = num_answers ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
cur_task . i_logits = i_logits ;
i_logits + = n_logits ;
2024-01-21 12:42:44 +00:00
n_cur + = cur_task . required_tokens ;
if ( + + i1 = = tasks . size ( ) ) {
break ;
}
}
if ( i0 = = i1 ) {
fprintf ( stderr , " %s : task %zu does not fit in the context window \n " , __func__ , i0 ) ;
return ;
}
llama_kv_cache_clear ( ctx ) ;
// decode all tasks [i0, i1)
if ( ! decode_helper ( ctx , batch , batch_logits , n_batch , n_vocab ) ) {
fprintf ( stderr , " %s: llama_decode() failed \n " , __func__ ) ;
return ;
}
// Compute log-probs in parallel
// First we collect all tasks
eval_pairs . clear ( ) ;
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & cur_task = tasks [ i ] ;
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
size_t li = 1 ; // skip the last logit of the common prefix (computed separately below)
2024-01-21 12:42:44 +00:00
for ( int s = 0 ; s < int ( cur_task . seq_tokens . size ( ) ) ; + + s ) {
for ( size_t j = cur_task . common_prefix ; j < cur_task . seq_tokens [ s ] . size ( ) - 1 ; j + + ) {
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
eval_pairs . emplace_back ( cur_task . i_logits + li + + , cur_task . seq_tokens [ s ] [ j + 1 ] ) ;
2024-01-21 12:42:44 +00:00
}
}
}
// Then we do the actual calculation
compute_logprobs ( batch_logits . data ( ) , n_vocab , workers , eval_pairs , eval_results ) ;
size_t ir = 0 ;
// compute the logprobs for each ending of the decoded tasks
for ( size_t i = i0 ; i < i1 ; + + i ) {
auto & cur_task = tasks [ i ] ;
//printf("==== Evaluating <%s> with correct answer ", cur_task.question.c_str());
//for (int j = 0; j < int(cur_task.mc1.labels.size()); ++j) {
// if (cur_task.mc1.labels[j] == 1) {
// printf("%d", j+1);
// }
//}
//printf("\n common_prefix: %zu\n", cur_task.common_prefix);
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// get the logits of the last token of the common prefix
std : : memcpy ( tok_logits . data ( ) , batch_logits . data ( ) + n_vocab * cur_task . i_logits , n_vocab * sizeof ( float ) ) ;
2024-01-21 12:42:44 +00:00
const auto first_probs = softmax ( tok_logits ) ;
cur_task . log_probs . resize ( cur_task . seq_tokens . size ( ) ) ;
for ( int s = 0 ; s < int ( cur_task . seq_tokens . size ( ) ) ; + + s ) {
size_t count = 1 ;
float log_prob = std : : log ( first_probs [ cur_task . seq_tokens [ s ] [ cur_task . common_prefix ] ] ) ;
for ( size_t j = cur_task . common_prefix ; j < cur_task . seq_tokens [ s ] . size ( ) - 1 ; j + + ) {
//printf(" %zu %g\n", ir, eval_results[ir]);
+ + count ;
log_prob + = eval_results [ ir + + ] ;
}
cur_task . log_probs [ s ] = log_prob / count ;
//printf(" Final: %g\n", log_prob / count);
//printf(" <%s> : %g\n", cur_task.mc1.answers[s].c_str(), log_prob/count);
}
// Find the ending with maximum logprob
size_t logprob_max_idx = 0 ;
float logprob_max_val = cur_task . log_probs [ 0 ] ;
for ( size_t s = 1 ; s < cur_task . log_probs . size ( ) ; s + + ) {
if ( cur_task . log_probs [ s ] > logprob_max_val ) {
logprob_max_val = cur_task . log_probs [ s ] ;
logprob_max_idx = s ;
}
}
n_tot_answers + = cur_task . log_probs . size ( ) ;
if ( cur_task . mc1 . labels [ logprob_max_idx ] = = 1 ) {
+ + n_correct ;
}
+ + n_done ;
// Print the accumulated accuracy mean x 100
printf ( " %d \t %.8lf \n " , n_done , 100. * n_correct / n_done ) ;
fflush ( stdout ) ;
}
i0 = i1 - 1 ;
}
llama_batch_free ( batch ) ;
2024-05-18 07:57:08 +00:00
if ( n_done < 100 & & ( params . multiple_choice_tasks ! = 0 & & params . multiple_choice_tasks < ( size_t ) n_task ) ) return ;
2024-01-21 12:42:44 +00:00
float p = 1.f * n_correct / n_done ;
float sigma = sqrt ( p * ( 1 - p ) / ( n_done - 1 ) ) ;
printf ( " \n Final result: %.4f +/- %.4f \n " , 100.f * p , 100.f * sigma ) ;
p = 1.f * n_done / n_tot_answers ;
sigma = sqrt ( p * ( 1 - p ) / ( n_done - 1 ) ) ;
printf ( " Random chance: %.4f +/- %.4f \n " , 100.f * p , 100.f * sigma ) ;
printf ( " \n " ) ;
}
2024-01-22 14:10:14 +00:00
static void kl_divergence ( llama_context * ctx , const gpt_params & params ) {
if ( params . logits_file . empty ( ) ) {
fprintf ( stderr , " %s: you must provide a name of a file containing the log probabilities of the base model \n " , __func__ ) ;
return ;
}
std : : ifstream in ( params . logits_file . c_str ( ) , std : : ios : : binary ) ;
if ( ! in ) {
fprintf ( stderr , " %s: failed to open %s \n " , __func__ , params . logits_file . c_str ( ) ) ;
return ;
}
{
char check [ 9 ] ; check [ 8 ] = 0 ;
in . read ( check , 8 ) ;
if ( in . fail ( ) | | strncmp ( " _logits_ " , check , 8 ) ! = 0 ) {
fprintf ( stderr , " %s: %s does not look like a file containing log-probabilities \n " , __func__ , params . logits_file . c_str ( ) ) ;
return ;
}
}
uint32_t n_ctx ;
in . read ( ( char * ) & n_ctx , sizeof ( n_ctx ) ) ;
if ( n_ctx > llama_n_ctx ( ctx ) ) {
2024-02-18 16:20:12 +00:00
fprintf ( stderr , " %s: %s has been computed with %u, while the current context is %d. Increase it with -c and retry \n " ,
2024-01-22 14:10:14 +00:00
__func__ , params . logits_file . c_str ( ) , n_ctx , params . n_ctx ) ;
}
int n_vocab , n_chunk ;
in . read ( ( char * ) & n_vocab , sizeof ( n_vocab ) ) ;
in . read ( ( char * ) & n_chunk , sizeof ( n_chunk ) ) ;
if ( in . fail ( ) ) {
2024-01-23 13:17:20 +00:00
fprintf ( stderr , " %s: failed reading n_vocab, n_chunk from %s \n " , __func__ , params . logits_file . c_str ( ) ) ;
2024-01-22 14:10:14 +00:00
return ;
}
if ( n_vocab ! = llama_n_vocab ( llama_get_model ( ctx ) ) ) {
fprintf ( stderr , " %s: inconsistent vocabulary (%d vs %d) \n " , __func__ , n_vocab , llama_n_vocab ( llama_get_model ( ctx ) ) ) ;
}
std : : vector < llama_token > tokens ( n_ctx * n_chunk ) ;
if ( in . read ( ( char * ) tokens . data ( ) , tokens . size ( ) * sizeof ( tokens [ 0 ] ) ) . fail ( ) ) {
fprintf ( stderr , " %s: failed reading evaluation tokens from %s \n " , __func__ , params . logits_file . c_str ( ) ) ;
return ;
}
const int n_batch = params . n_batch ;
const int num_batches = ( n_ctx + n_batch - 1 ) / n_batch ;
const int nv = 2 * ( ( n_vocab + 1 ) / 2 ) + 4 ;
2024-08-15 07:23:23 +00:00
const bool add_bos = llama_add_bos_token ( llama_get_model ( ctx ) ) ;
GGML_ASSERT ( ! llama_add_eos_token ( llama_get_model ( ctx ) ) ) ;
2024-01-22 14:10:14 +00:00
std : : vector < uint16_t > log_probs_uint16 ( size_t ( n_ctx - 1 - n_ctx / 2 ) * nv ) ;
2024-04-30 21:36:27 +00:00
std : : vector < float > kld_values ( size_t ( n_ctx - 1 - n_ctx / 2 ) * n_chunk ) ;
std : : vector < float > p_diff_values ( size_t ( n_ctx - 1 - n_ctx / 2 ) * n_chunk ) ;
2024-01-22 14:10:14 +00:00
std : : vector < float > logits ;
if ( num_batches > 1 ) {
logits . reserve ( n_ctx * n_vocab ) ;
}
std : : vector < std : : thread > workers ( std : : thread : : hardware_concurrency ( ) - 1 ) ;
auto mean_and_uncertainty = [ ] ( double sum , double sum2 , size_t count ) {
if ( count < 1 ) {
return std : : make_pair ( 0. , 0. ) ;
}
double f = sum / count ;
double df = sum2 / count - f * f ;
df = df > 0 & & count > 10 ? sqrt ( df / ( count - 1 ) ) : 0. ;
return std : : make_pair ( f , df ) ;
} ;
2024-04-30 21:36:27 +00:00
auto covariance = [ ] ( double suma , double sumb , double sumab , size_t count ) {
if ( count < 10 ) {
return 0.0 ;
}
double var = sumab / count - ( suma / count ) * ( sumb / count ) ;
var / = count - 1 ;
return var ;
} ;
2024-01-22 14:10:14 +00:00
kl_divergence_result kld ;
2024-04-30 21:36:27 +00:00
auto kld_ptr = kld_values . data ( ) ;
auto p_diff_ptr = p_diff_values . data ( ) ;
2024-01-22 14:10:14 +00:00
for ( int i = 0 ; i < n_chunk ; + + i ) {
const int start = i * n_ctx ;
const int end = start + n_ctx ;
const auto t_start = std : : chrono : : high_resolution_clock : : now ( ) ;
if ( in . read ( ( char * ) log_probs_uint16 . data ( ) , log_probs_uint16 . size ( ) * sizeof ( uint16_t ) ) . fail ( ) ) {
fprintf ( stderr , " %s: failed reading log-probs for chunk %d \n " , __func__ , i ) ;
return ;
}
// clear the KV cache
llama_kv_cache_clear ( ctx ) ;
for ( int j = 0 ; j < num_batches ; + + j ) {
const int batch_start = start + j * n_batch ;
const int batch_size = std : : min ( end - batch_start , n_batch ) ;
// save original token and restore it after eval
const auto token_org = tokens [ batch_start ] ;
// add BOS token for the first batch of each chunk
if ( add_bos & & j = = 0 ) {
tokens [ batch_start ] = llama_token_bos ( llama_get_model ( ctx ) ) ;
}
llama : greatly reduce output buffer memory usage (#6122)
* llama : greatly reduce logits memory usage
* llama : more compact state saving and reloading
* llama : fix lctx.n_outputs not being set before building graph
* perplexity : adapt to the logits API changes
* perplexity : fix Winogrande, use correct logits for second choice start
The first logits used to evaluate the second choice were not from
the end of the common prefix; instead, they were the logits from the end
of the first choice. This has been corrected.
The previous implementation sometimes had outliers in the scores of
choices for some tasks, and the logic to skip choices words
in the log-likelihood evaluation probably was an attempt to reduce those,
but it was complex and didn't quite seem to be the right thing.
This is simpler now, and the outlier scores aren't there anymore.
* perplexity : normalize spaces and punctuation in Winogrande sentences
* llama : fix embedding conditions
* llama : fix llama_get_embeddings_ith when the resulting id is 0
* llama : fix wrong n_outputs in llama_set_inputs
A mismatch happened when using a smaller n_ubatch than n_batch and then using
llama_batch_get_one(). The decision of what n_outputs should be now almost
fully depends on how lctx.n_outputs is set in llama_decode_internal.
The conditions are simpler this way.
* llama : when saving the state, recalculate n_outputs
This ensures the correct number of outputs for the entire previous batch
is stored in the session file, even when n_ubatch is smaller than n_batch.
* llama : fix not-skipping outputs of non-causal models
* llama : fix running a batch with n_outputs == 0
It previously worked because lctx.inp_out_ids was not initialized,
so it pointed to some garbage address which was somehow still valid when I
ran my tests.
* llama : keep same graph topology even when n_outputs == 0
* ggml : saner ggml_can_repeat with empty tensors
* ggml : future-proof ggml_is_empty by using GGML_MAX_DIMS - 1
* ggml : do not multi-thread ops returning empty tensors
* ggml : make ggml_is_empty public and work with views
* llama : use a vector for ctx->output_ids
* llama : rework reallocation logic for llama_output_reserve
Now comparing the actual size with the new total size of the output buffer
to allow more efficient enabling and disabling of the embeddings
and/or logits output in the future.
* ggml : skip empty tensors in all backends
* llama : fix llama_output_reserve nullptr deref when new_size is 0
* perplexity : make Winogrande work as it does on master
The problems with the Winogrande implementation will
need to be fixed in a separate PR to ease review.
* llama : clearer error messages for invalid logits or embeddings ids
* llama : assert all models that can have inp_out_ids
Since the graph topology is now constant, this presence check
can be done even when there are no outputs.
* llama : assert logits and embd buffers exist before writing to them
* llama : handle errors from llama_output_reserve at call sites
* perplexity : make hellaswag and multiple-choice outputs identical to master
Due to how the KV cache is updated, the logprobs for tokens in a batch
are very slightly affected by the other tokens present in the batch,
so to make hellaswag and multiple-choice return exactly the same results
as on master, the last token of each sequence needs to be evaluated
even though its output is not used at all.
This will probably be changed back in the future to make these benchmarks
a tiny bit faster.
* perplexity : fix division by zero when using less than 100 multiple-choice tasks
* llama : allow loading state saved with a different ctx size
When loading a session file, the context size is now only required to be
at least enough to load the KV cells contained in that session file,
instead of requiring to use exactly the same context size as when saving.
Doing this enables the use-case of extending or shrinking the context size
of a saved session.
This breaks existing session files because the meaning of kv_buf_size
is slightly changed (previously it was the size of the whole KV cache,
now it's only the size of the saved part of it). This allows for
finer-grained sanity checks when loading in an effort to keep kv_buf_size
useful even when the kv_size is changed.
* llama : minor
ggml-ci
* readme : update recent API changes, and warn about Vulkan
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2024-03-26 14:46:41 +00:00
// TODO: use llama_batch.logits instead of relying on logits_all == true
2024-01-22 14:10:14 +00:00
if ( llama_decode ( ctx , llama_batch_get_one ( tokens . data ( ) + batch_start , batch_size , j * n_batch , 0 ) ) ) {
fprintf ( stderr , " %s : failed to eval \n " , __func__ ) ;
return ;
}
// restore the original token in case it was set to BOS
tokens [ batch_start ] = token_org ;
if ( num_batches > 1 ) {
const auto * batch_logits = llama_get_logits ( ctx ) ;
logits . insert ( logits . end ( ) , batch_logits , batch_logits + batch_size * n_vocab ) ;
}
}
const auto t_end = std : : chrono : : high_resolution_clock : : now ( ) ;
if ( i = = 0 ) {
const float t_total = std : : chrono : : duration < float > ( t_end - t_start ) . count ( ) ;
fprintf ( stderr , " %s: %.2f seconds per pass - ETA " , __func__ , t_total ) ;
int total_seconds = ( int ) ( t_total * n_chunk ) ;
if ( total_seconds > = 60 * 60 ) {
fprintf ( stderr , " %d hours " , total_seconds / ( 60 * 60 ) ) ;
total_seconds = total_seconds % ( 60 * 60 ) ;
}
fprintf ( stderr , " %.2f minutes \n " , total_seconds / 60.0 ) ;
2024-04-30 21:36:27 +00:00
printf ( " \n chunk PPL ln(PPL(Q)/PPL(base)) KL Divergence Δp RMS Same top p \n " ) ;
2024-01-22 14:10:14 +00:00
}
const int first = n_ctx / 2 ;
const float * all_logits = num_batches > 1 ? logits . data ( ) : llama_get_logits ( ctx ) ;
process_logits ( n_vocab , all_logits + first * n_vocab , tokens . data ( ) + start + first , n_ctx - 1 - first ,
2024-04-30 21:36:27 +00:00
workers , log_probs_uint16 , kld , kld_ptr , p_diff_ptr ) ;
p_diff_ptr + = n_ctx - 1 - first ;
kld_ptr + = n_ctx - 1 - first ;
printf ( " %4d " , i + 1 ) ;
auto log_ppl = mean_and_uncertainty ( kld . sum_nll , kld . sum_nll2 , kld . count ) ;
const double ppl_val = exp ( log_ppl . first ) ;
const double ppl_unc = ppl_val * log_ppl . second ; // ppl_unc = sqrt( (dexp(x) / dx) ** 2 * log_ppl.second ** 2 )
printf ( " %9.4lf ± %9.4lf " , ppl_val , ppl_unc ) ;
auto log_ppl_base = mean_and_uncertainty ( kld . sum_nll_base , kld . sum_nll_base2 , kld . count ) ;
const double log_ppl_cov = covariance ( kld . sum_nll , kld . sum_nll_base , kld . sum_nll_nll_base , kld . count ) ;
const double log_ppl_ratio_val = log_ppl . first - log_ppl_base . first ;
const double log_ppl_ratio_unc = sqrt ( log_ppl . second * log_ppl . second + log_ppl_base . second * log_ppl_base . second - 2.0 * log_ppl_cov ) ;
printf ( " %10.5lf ± %10.5lf " , log_ppl_ratio_val , log_ppl_ratio_unc ) ;
auto kl_div = mean_and_uncertainty ( kld . sum_kld , kld . sum_kld2 , kld . count ) ;
printf ( " %10.5lf ± %10.5lf " , kl_div . first , kl_div . second ) ;
2024-01-22 14:10:14 +00:00
2024-04-30 21:36:27 +00:00
auto p_diff_mse = mean_and_uncertainty ( kld . sum_p_diff2 , kld . sum_p_diff4 , kld . count ) ;
const double p_diff_rms_val = sqrt ( p_diff_mse . first ) ;
const double p_diff_rms_unc = 0.5 / p_diff_rms_val * p_diff_mse . second ;
printf ( " %6.3lf ± %6.3lf %% " , 100.0 * p_diff_rms_val , 100.0 * p_diff_rms_unc ) ;
2024-01-22 14:10:14 +00:00
2024-04-30 21:36:27 +00:00
double p_top_val = 1. * kld . n_same_top / kld . count ;
double p_top_unc = sqrt ( p_top_val * ( 1 - p_top_val ) / ( kld . count - 1 ) ) ;
printf ( " %6.3lf ± %6.3lf %% " , 100.0 * p_top_val , 100.0 * p_top_unc ) ;
printf ( " \n " ) ;
2024-01-22 14:10:14 +00:00
fflush ( stdout ) ;
logits . clear ( ) ;
}
printf ( " \n " ) ;
2024-01-23 13:17:20 +00:00
if ( kld . count < 100 ) return ; // we do not wish to do statistics on so few values
std : : sort ( kld_values . begin ( ) , kld_values . end ( ) ) ;
2024-04-30 21:36:27 +00:00
std : : sort ( p_diff_values . begin ( ) , p_diff_values . end ( ) ) ;
printf ( " ====== Perplexity statistics ====== \n " ) ;
auto log_ppl = mean_and_uncertainty ( kld . sum_nll , kld . sum_nll2 , kld . count ) ;
const double ppl_val = exp ( log_ppl . first ) ;
const double ppl_unc = ppl_val * log_ppl . second ; // ppl_unc = sqrt( (dexp(x) / dx) ** 2 * log_ppl.second ** 2 )
printf ( " Mean PPL(Q) : %10.6lf ± %10.6lf \n " , ppl_val , ppl_unc ) ;
auto log_ppl_base = mean_and_uncertainty ( kld . sum_nll_base , kld . sum_nll_base2 , kld . count ) ;
const double ppl_base_val = exp ( log_ppl_base . first ) ;
const double ppl_base_unc = ppl_base_val * log_ppl_base . second ; // ppl_base_unc = sqrt( (dexp(x) / dx) ** 2 * log_ppl_base.second ** 2 )
printf ( " Mean PPL(base) : %10.6lf ± %10.6lf \n " , ppl_base_val , ppl_base_unc ) ;
const double log_ppl_cov = covariance ( kld . sum_nll , kld . sum_nll_base , kld . sum_nll_nll_base , kld . count ) ;
// printf("Cov(ln(PPL(Q)), ln(PPL(base))): %10.6lf\n", log_ppl_cov);
const double log_ppl_cor = log_ppl_cov / ( log_ppl . second * log_ppl_base . second ) ;
printf ( " Cor(ln(PPL(Q)), ln(PPL(base))): %6.2lf%% \n " , 100.0 * log_ppl_cor ) ;
const double log_ppl_ratio_val = log_ppl . first - log_ppl_base . first ;
const double log_ppl_ratio_unc = sqrt ( log_ppl . second * log_ppl . second + log_ppl_base . second * log_ppl_base . second - 2.0 * log_ppl_cov ) ;
printf ( " Mean ln(PPL(Q)/PPL(base)) : %10.6lf ± %10.6lf \n " , log_ppl_ratio_val , log_ppl_ratio_unc ) ;
2024-01-23 13:17:20 +00:00
2024-04-30 21:36:27 +00:00
const double ppl_ratio_val = exp ( log_ppl_ratio_val ) ;
const double ppl_ratio_unc = ppl_ratio_val * log_ppl_ratio_unc ; // ppl_ratio_unc = sqrt( (dexp(x) / dx) ** 2 * log_ppl_ratio.second ** 2 )
printf ( " Mean PPL(Q)/PPL(base) : %10.6lf ± %10.6lf \n " , ppl_ratio_val , ppl_ratio_unc ) ;
const double ppl_cov = ppl_val * ppl_base_val * log_ppl_cov ;
const double ppl_diff_val = ppl_val - ppl_base_val ;
const double ppl_diff_unc = sqrt ( ppl_unc * ppl_unc + ppl_base_unc * ppl_base_unc - 2.0 * ppl_cov ) ;
printf ( " Mean PPL(Q)-PPL(base) : %10.6lf ± %10.6lf \n " , ppl_diff_val , ppl_diff_unc ) ;
printf ( " \n " ) ;
printf ( " ====== KL divergence statistics ====== \n " ) ;
2024-01-23 13:17:20 +00:00
auto kl_div = mean_and_uncertainty ( kld . sum_kld , kld . sum_kld2 , kld . count ) ;
2024-04-30 21:36:27 +00:00
printf ( " Mean KLD: %10.6lf ± %10.6lf \n " , kl_div . first , kl_div . second ) ;
2024-01-23 13:17:20 +00:00
auto kld_median = kld_values . size ( ) % 2 = = 0 ? 0.5f * ( kld_values [ kld_values . size ( ) / 2 ] + kld_values [ kld_values . size ( ) / 2 - 1 ] )
: kld_values [ kld_values . size ( ) / 2 ] ;
2024-04-30 21:36:27 +00:00
auto percentile = [ ] ( std : : vector < float > values , float fraction ) {
if ( fraction < = 0 ) return values . front ( ) ;
if ( fraction > = 1 ) return values . back ( ) ;
float p = fraction * ( values . size ( ) - 1 ) ;
2024-01-23 13:17:20 +00:00
size_t ip = size_t ( p ) ; p - = ip ;
2024-04-30 21:36:27 +00:00
return ( 1 - p ) * values [ ip ] + p * values [ std : : min ( ip + 1 , values . size ( ) - 1 ) ] ;
2024-01-23 13:17:20 +00:00
} ;
2024-04-30 21:36:27 +00:00
printf ( " Maximum KLD: %10.6f \n " , kld_values . back ( ) ) ;
printf ( " 99.9%% KLD: %10.6f \n " , percentile ( kld_values , 0.999f ) ) ;
printf ( " 99.0%% KLD: %10.6f \n " , percentile ( kld_values , 0.990f ) ) ;
printf ( " 99.0%% KLD: %10.6f \n " , percentile ( kld_values , 0.990f ) ) ;
printf ( " Median KLD: %10.6f \n " , kld_median ) ;
printf ( " 10.0%% KLD: %10.6f \n " , percentile ( kld_values , 0.100f ) ) ;
printf ( " 5.0%% KLD: %10.6f \n " , percentile ( kld_values , 0.050f ) ) ;
printf ( " 1.0%% KLD: %10.6f \n " , percentile ( kld_values , 0.010f ) ) ;
printf ( " Minimum KLD: %10.6f \n " , kld_values . front ( ) ) ;
printf ( " \n " ) ;
2024-01-23 13:17:20 +00:00
2024-04-30 21:36:27 +00:00
printf ( " ====== Token probability statistics ====== \n " ) ;
auto p_diff = mean_and_uncertainty ( kld . sum_p_diff , kld . sum_p_diff2 , kld . count ) ;
printf ( " Mean Δp: %6.3lf ± %5.3lf %% \n " , 100.0 * p_diff . first , 100.0 * p_diff . second ) ;
auto p_diff_median = p_diff_values . size ( ) % 2 = = 0 ? 0.5f * ( p_diff_values [ p_diff_values . size ( ) / 2 ] + p_diff_values [ p_diff_values . size ( ) / 2 - 1 ] )
: p_diff_values [ p_diff_values . size ( ) / 2 ] ;
printf ( " Maximum Δp: %6.3lf%% \n " , 100.0 * p_diff_values . back ( ) ) ;
printf ( " 99.9%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.999f ) ) ;
printf ( " 99.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.990f ) ) ;
printf ( " 95.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.950f ) ) ;
printf ( " 90.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.900f ) ) ;
printf ( " 75.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.750f ) ) ;
printf ( " Median Δp: %6.3lf%% \n " , 100.0 * p_diff_median ) ;
printf ( " 25.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.250f ) ) ;
printf ( " 10.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.100f ) ) ;
printf ( " 5.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.050f ) ) ;
printf ( " 1.0%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.010f ) ) ;
printf ( " 0.1%% Δp: %6.3lf%% \n " , 100.0 * percentile ( p_diff_values , 0.001f ) ) ;
printf ( " Minimum Δp: %6.3lf%% \n " , 100.0 * p_diff_values . front ( ) ) ;
auto p_diff_mse = mean_and_uncertainty ( kld . sum_p_diff2 , kld . sum_p_diff4 , kld . count ) ;
// printf("MSE Δp : %10.6lf ± %10.6lf\n", p_diff_mse.first, p_diff_mse.second);
const double p_diff_rms_val = sqrt ( p_diff_mse . first ) ;
const double p_diff_rms_unc = 0.5 / p_diff_rms_val * p_diff_mse . second ;
printf ( " RMS Δp : %6.3lf ± %5.3lf %% \n " , 100.0 * p_diff_rms_val , 100.0 * p_diff_rms_unc ) ;
const double same_top_p = 1.0 * kld . n_same_top / kld . count ;
printf ( " Same top p: %6.3lf ± %5.3lf %% \n " , 100.0 * same_top_p , 100.0 * sqrt ( same_top_p * ( 1.0 - same_top_p ) / ( kld . count - 1 ) ) ) ;
2024-01-23 13:17:20 +00:00
2024-01-22 14:10:14 +00:00
}
2024-01-18 11:46:27 +00:00
2023-03-25 18:26:40 +00:00
int main ( int argc , char * * argv ) {
gpt_params params ;
2024-06-04 18:23:39 +00:00
params . n_ctx = 512 ;
params . logits_all = true ;
2024-09-09 21:36:09 +00:00
if ( ! gpt_params_parse ( argc , argv , params , LLAMA_EXAMPLE_PERPLEXITY ) ) {
2023-03-25 18:26:40 +00:00
return 1 ;
}
2024-03-09 18:55:54 +00:00
const int32_t n_ctx = params . n_ctx ;
2024-04-16 06:28:33 +00:00
if ( n_ctx < = 0 ) {
fprintf ( stderr , " %s: perplexity tool requires '--ctx-size' > 0 \n " , __func__ ) ;
return 1 ;
}
2024-03-09 18:55:54 +00:00
const bool ppl = ! params . hellaswag & & ! params . winogrande & & ! params . multiple_choice & & ! params . kl_divergence ;
2024-04-16 06:28:33 +00:00
2024-03-09 18:55:54 +00:00
if ( ppl ) {
2024-04-16 06:28:33 +00:00
const int32_t n_seq = std : : max ( 1 , params . n_batch / n_ctx ) ;
const int32_t n_kv = n_seq * n_ctx ;
2024-03-09 18:55:54 +00:00
params . n_parallel = n_seq ;
2024-04-16 06:28:33 +00:00
params . n_ctx = n_kv ;
2024-03-09 18:55:54 +00:00
params . n_batch = std : : min ( params . n_batch , n_kv ) ;
} else {
params . n_batch = std : : min ( params . n_batch , params . n_ctx ) ;
2024-07-03 17:33:31 +00:00
if ( params . kl_divergence ) {
params . n_parallel = 1 ;
} else {
// ensure there's at least enough seq_ids for HellaSwag
params . n_parallel = std : : max ( 4 , params . n_parallel ) ;
}
2024-03-09 18:55:54 +00:00
}
2023-03-25 18:26:40 +00:00
2023-08-23 09:56:42 +00:00
if ( params . ppl_stride > 0 ) {
fprintf ( stderr , " Will perform strided perplexity calculation -> adjusting context size from %d to %d \n " ,
params . n_ctx , params . n_ctx + params . ppl_stride / 2 ) ;
params . n_ctx + = params . ppl_stride / 2 ;
}
2023-09-15 20:59:49 +00:00
print_build_info ( ) ;
2023-05-01 16:23:47 +00:00
2024-02-16 09:31:07 +00:00
llama_backend_init ( ) ;
llama_numa_init ( params . numa ) ;
2023-05-20 08:06:11 +00:00
2023-05-02 20:39:51 +00:00
// load the model and apply lora adapter, if any
2024-08-05 16:14:10 +00:00
llama_init_result llama_init = llama_init_from_gpt_params ( params ) ;
llama_model * model = llama_init . model ;
llama_context * ctx = llama_init . context ;
2023-06-24 08:47:58 +00:00
if ( model = = NULL ) {
2023-05-02 20:39:51 +00:00
fprintf ( stderr , " %s: error: unable to load model \n " , __func__ ) ;
return 1 ;
2023-04-17 15:28:55 +00:00
}
2023-09-28 19:42:38 +00:00
const int n_ctx_train = llama_n_ctx_train ( model ) ;
2024-06-04 18:23:39 +00:00
2023-09-08 15:43:35 +00:00
if ( params . n_ctx > n_ctx_train ) {
fprintf ( stderr , " %s: warning: model was trained on only %d context tokens (%d specified) \n " ,
__func__ , n_ctx_train , params . n_ctx ) ;
2023-09-04 19:26:24 +00:00
}
2023-03-25 18:26:40 +00:00
// print system information
{
fprintf ( stderr , " \n " ) ;
2024-05-22 17:04:20 +00:00
fprintf ( stderr , " %s \n " , gpt_params_get_system_info ( params ) . c_str ( ) ) ;
2023-03-25 18:26:40 +00:00
}
2023-08-28 15:59:39 +00:00
struct results_perplexity results ;
2023-07-28 18:25:36 +00:00
if ( params . hellaswag ) {
hellaswag_score ( ctx , params ) ;
2024-01-18 11:46:27 +00:00
} else if ( params . winogrande ) {
winogrande_score ( ctx , params ) ;
2024-01-21 12:42:44 +00:00
} else if ( params . multiple_choice ) {
multiple_choice_score ( ctx , params ) ;
2024-01-22 14:10:14 +00:00
} else if ( params . kl_divergence ) {
kl_divergence ( ctx , params ) ;
2023-07-22 12:21:24 +00:00
} else {
2024-03-09 18:55:54 +00:00
results = perplexity ( ctx , params , n_ctx ) ;
2023-07-22 12:21:24 +00:00
}
2023-03-25 18:26:40 +00:00
2024-09-07 12:16:19 +00:00
LOG_TEE ( " \n " ) ;
2024-09-13 06:53:38 +00:00
llama_perf_context_print ( ctx ) ;
2023-08-28 15:59:39 +00:00
write_logfile ( ctx , params , model , results ) ;
2023-03-25 18:26:40 +00:00
llama_free ( ctx ) ;
2023-06-24 08:47:58 +00:00
llama_free_model ( model ) ;
2023-03-25 18:26:40 +00:00
2023-07-10 15:49:56 +00:00
llama_backend_free ( ) ;
2023-03-25 18:26:40 +00:00
return 0 ;
}