mirror of
https://github.com/ggerganov/llama.cpp.git
synced 2024-11-14 14:59:52 +00:00
1b9ae5189c
* (wip) argparser v3 * migrated * add test * handle env * fix linux build * add export-docs example * fix build (2) * skip build test-arg-parser on windows * update server docs * bring back missing --alias * bring back --n-predict * clarify test-arg-parser * small correction * add comments * fix args with 2 values * refine example-specific args * no more lamba capture Co-authored-by: slaren@users.noreply.github.com * params.sparams * optimize more * export-docs --> gen-docs
257 lines
8.0 KiB
C++
257 lines
8.0 KiB
C++
#include "common.h"
|
|
#include "llama.h"
|
|
|
|
#include <vector>
|
|
#include <cstdio>
|
|
|
|
int main(int argc, char ** argv) {
|
|
gpt_params params;
|
|
|
|
params.prompt = "The quick brown fox";
|
|
params.sparams.seed = 1234;
|
|
|
|
auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
|
|
if (!gpt_params_parse(argc, argv, params, options)) {
|
|
return 1;
|
|
}
|
|
|
|
print_build_info();
|
|
|
|
if (params.n_predict < 0) {
|
|
params.n_predict = 16;
|
|
}
|
|
|
|
auto n_past = 0;
|
|
|
|
std::string result0;
|
|
std::string result1;
|
|
std::string result2;
|
|
|
|
// init
|
|
llama_init_result llama_init = llama_init_from_gpt_params(params);
|
|
|
|
llama_model * model = llama_init.model;
|
|
llama_context * ctx = llama_init.context;
|
|
|
|
if (model == nullptr || ctx == nullptr) {
|
|
fprintf(stderr, "%s : failed to init\n", __func__);
|
|
return 1;
|
|
}
|
|
|
|
auto sparams = llama_sampler_chain_default_params();
|
|
|
|
llama_sampler * smpl = llama_sampler_chain_init(sparams);
|
|
|
|
llama_sampler_chain_add(smpl, llama_sampler_init_softmax());
|
|
llama_sampler_chain_add(smpl, llama_sampler_init_dist(params.sparams.seed));
|
|
|
|
// tokenize prompt
|
|
auto tokens = llama_tokenize(ctx, params.prompt, true);
|
|
|
|
// evaluate prompt
|
|
llama_decode(ctx, llama_batch_get_one(tokens.data(), tokens.size(), n_past, 0));
|
|
n_past += tokens.size();
|
|
|
|
// save state (rng, logits, embedding and kv_cache) to file
|
|
{
|
|
std::vector<uint8_t> state_mem(llama_state_get_size(ctx));
|
|
const size_t written = llama_state_get_data(ctx, state_mem.data(), state_mem.size());
|
|
|
|
FILE *fp_write = fopen("dump_state.bin", "wb");
|
|
fwrite(state_mem.data(), 1, written, fp_write);
|
|
fclose(fp_write);
|
|
|
|
fprintf(stderr, "%s : serialized state into %zd out of a maximum of %zd bytes\n", __func__, written, state_mem.size());
|
|
}
|
|
|
|
// save state (last tokens)
|
|
const auto n_past_saved = n_past;
|
|
|
|
// first run
|
|
printf("\nfirst run: %s", params.prompt.c_str());
|
|
|
|
for (auto i = 0; i < params.n_predict; i++) {
|
|
auto next_token = llama_sampler_sample(smpl, ctx, -1);
|
|
auto next_token_str = llama_token_to_piece(ctx, next_token);
|
|
|
|
llama_sampler_accept(smpl, next_token);
|
|
|
|
printf("%s", next_token_str.c_str());
|
|
result0 += next_token_str;
|
|
|
|
if (llama_decode(ctx, llama_batch_get_one(&next_token, 1, n_past, 0))) {
|
|
fprintf(stderr, "\n%s : failed to evaluate\n", __func__);
|
|
llama_free(ctx);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
n_past += 1;
|
|
}
|
|
|
|
printf("\n\n");
|
|
|
|
// free old context
|
|
llama_free(ctx);
|
|
|
|
// make new context
|
|
auto * ctx2 = llama_new_context_with_model(model, llama_context_params_from_gpt_params(params));
|
|
|
|
llama_sampler * smpl2 = llama_sampler_chain_init(sparams);
|
|
|
|
llama_sampler_chain_add(smpl2, llama_sampler_init_softmax());
|
|
llama_sampler_chain_add(smpl2, llama_sampler_init_dist(params.sparams.seed));
|
|
|
|
printf("\nsecond run: %s", params.prompt.c_str());
|
|
|
|
// load state (rng, logits, embedding and kv_cache) from file
|
|
{
|
|
std::vector<uint8_t> state_mem;
|
|
|
|
FILE * fp_read = fopen("dump_state.bin", "rb");
|
|
fseek(fp_read, 0, SEEK_END);
|
|
state_mem.resize(ftell(fp_read));
|
|
fseek(fp_read, 0, SEEK_SET);
|
|
const size_t read = fread(state_mem.data(), 1, state_mem.size(), fp_read);
|
|
fclose(fp_read);
|
|
|
|
if (read != llama_state_set_data(ctx2, state_mem.data(), state_mem.size())) {
|
|
fprintf(stderr, "\n%s : failed to read state\n", __func__);
|
|
llama_free(ctx2);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
|
|
fprintf(stderr, "%s : deserialized state from %zd out of a maximum of %zd bytes\n", __func__, read, state_mem.size());
|
|
}
|
|
|
|
// restore state (last tokens)
|
|
n_past = n_past_saved;
|
|
|
|
// second run
|
|
for (auto i = 0; i < params.n_predict; i++) {
|
|
auto next_token = llama_sampler_sample(smpl2, ctx2, -1);
|
|
auto next_token_str = llama_token_to_piece(ctx2, next_token);
|
|
|
|
llama_sampler_accept(smpl2, next_token);
|
|
|
|
printf("%s", next_token_str.c_str());
|
|
result1 += next_token_str;
|
|
|
|
if (llama_decode(ctx2, llama_batch_get_one(&next_token, 1, n_past, 0))) {
|
|
fprintf(stderr, "\n%s : failed to evaluate\n", __func__);
|
|
llama_free(ctx2);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
n_past += 1;
|
|
}
|
|
|
|
printf("\n\n");
|
|
|
|
llama_free(ctx2);
|
|
|
|
if (result0 != result1) {
|
|
fprintf(stderr, "\n%s : error : the 2 generations are different\n", __func__);
|
|
return 1;
|
|
}
|
|
|
|
// make new context
|
|
auto * ctx3 = llama_new_context_with_model(model, llama_context_params_from_gpt_params(params));
|
|
|
|
llama_sampler * smpl3 = llama_sampler_chain_init(sparams);
|
|
|
|
llama_sampler_chain_add(smpl3, llama_sampler_init_softmax());
|
|
llama_sampler_chain_add(smpl3, llama_sampler_init_dist(params.sparams.seed));
|
|
|
|
printf("\nsingle seq run: %s", params.prompt.c_str());
|
|
|
|
// load state (rng, logits, embedding and kv_cache) from file
|
|
{
|
|
std::vector<uint8_t> state_mem;
|
|
|
|
FILE * fp_read = fopen("dump_state.bin", "rb");
|
|
fseek(fp_read, 0, SEEK_END);
|
|
state_mem.resize(ftell(fp_read));
|
|
fseek(fp_read, 0, SEEK_SET);
|
|
const size_t read = fread(state_mem.data(), 1, state_mem.size(), fp_read);
|
|
fclose(fp_read);
|
|
|
|
if (read != llama_state_set_data(ctx3, state_mem.data(), state_mem.size())) {
|
|
fprintf(stderr, "\n%s : failed to read state\n", __func__);
|
|
llama_free(ctx3);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
|
|
fprintf(stderr, "%s : deserialized state from %zd out of a maximum of %zd bytes\n", __func__, read, state_mem.size());
|
|
}
|
|
|
|
// restore state (last tokens)
|
|
n_past = n_past_saved;
|
|
|
|
// save seq 0 and load into seq 1
|
|
{
|
|
// save kv of seq 0
|
|
std::vector<uint8_t> seq_store(llama_state_seq_get_size(ctx3, 0));
|
|
const size_t ncopy = llama_state_seq_get_data(ctx3, seq_store.data(), seq_store.size(), 0);
|
|
if (ncopy != seq_store.size()) {
|
|
fprintf(stderr, "\n%s : seq copy data length %zd does not match expected length %zd\n", __func__, ncopy, seq_store.size());
|
|
llama_free(ctx3);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
fprintf(stderr, "%s : seq 0 copied, %zd bytes\n", __func__, ncopy);
|
|
|
|
// erase whole kv
|
|
llama_kv_cache_clear(ctx3);
|
|
fprintf(stderr, "%s : kv cache cleared\n", __func__);
|
|
|
|
// restore kv into seq 1
|
|
const size_t nset = llama_state_seq_set_data(ctx3, seq_store.data(), seq_store.size(), 1);
|
|
if (nset != seq_store.size()) {
|
|
fprintf(stderr, "\n%s : seq set data length %zd does not match expected length %zd\n", __func__, nset, seq_store.size());
|
|
llama_free(ctx3);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
fprintf(stderr, "%s : seq 1 restored, %zd bytes\n", __func__, nset);
|
|
}
|
|
|
|
// third run with seq 1 instead of 0
|
|
for (auto i = 0; i < params.n_predict; i++) {
|
|
auto next_token = llama_sampler_sample(smpl3, ctx3, -1);
|
|
auto next_token_str = llama_token_to_piece(ctx3, next_token);
|
|
|
|
llama_sampler_accept(smpl3, next_token);
|
|
|
|
printf("%s", next_token_str.c_str());
|
|
result2 += next_token_str;
|
|
|
|
if (llama_decode(ctx3, llama_batch_get_one(&next_token, 1, n_past, 1))) {
|
|
fprintf(stderr, "\n%s : failed to evaluate\n", __func__);
|
|
llama_free(ctx3);
|
|
llama_free_model(model);
|
|
return 1;
|
|
}
|
|
n_past += 1;
|
|
}
|
|
|
|
printf("\n");
|
|
|
|
llama_sampler_free(smpl);
|
|
llama_sampler_free(smpl2);
|
|
llama_sampler_free(smpl3);
|
|
|
|
llama_free(ctx3);
|
|
llama_free_model(model);
|
|
|
|
if (result0 != result2) {
|
|
fprintf(stderr, "\n%s : error : the seq restore generation is different\n", __func__);
|
|
return 1;
|
|
}
|
|
|
|
fprintf(stderr, "\n%s : success\n", __func__);
|
|
|
|
return 0;
|
|
}
|