root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-10 18:51:45 +00:00
Code Issues Actions 13 Packages Projects Releases Wiki Activity

common : sanity check for non-NULL tokens

Browse Source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-10-04 12:04:54 +03:00
parent 1ba3df3de5
commit 9e897d4439
No known key found for this signature in database
GPG Key ID: BF970631944C16B7
3 changed files with 39 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
common/common.cpp
View File

@ -838,6 +838,31 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
return iparams; return iparams;
} }
if (params.reranking) {
bool ok = true;
if (llama_token_bos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have a BOS token, reranking will not work\n", __func__);
ok = false;
}
if (llama_token_eos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have an EOS token, reranking will not work\n", __func__);
ok = false;
}
if (llama_token_sep(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have a SEP token, reranking will not work\n", __func__);
ok = false;
}
if (!ok) {
llama_free_model(model);
return iparams;
}
}
auto cparams = llama_context_params_from_gpt_params(params); auto cparams = llama_context_params_from_gpt_params(params);
llama_context * lctx = llama_new_context_with_model(model, cparams); llama_context * lctx = llama_new_context_with_model(model, cparams);
@ -855,6 +880,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
if (cvec.n_embd == -1) { if (cvec.n_embd == -1) {
llama_free(lctx); llama_free(lctx);
llama_free_model(model); llama_free_model(model);
return iparams; return iparams;
} }
@ -867,6 +893,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
if (err) { if (err) {
llama_free(lctx); llama_free(lctx);
llama_free_model(model); llama_free_model(model);
return iparams; return iparams;
} }
} }
@ -889,7 +916,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
llama_lora_adapters_apply(lctx, iparams.lora_adapters); llama_lora_adapters_apply(lctx, iparams.lora_adapters);
} }
if (params.sparams.ignore_eos && llama_token_eos(model) == -1) { if (params.sparams.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__); LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__);
params.sparams.ignore_eos = false; params.sparams.ignore_eos = false;
} }
@ -930,6 +957,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
iparams.model = model; iparams.model = model;
iparams.context = lctx; iparams.context = lctx;
return iparams; return iparams;
} }

18
src/llama-vocab.h
View File

@ -40,17 +40,17 @@ struct llama_vocab {
id special_bos_id = 1; id special_bos_id = 1;
id special_eos_id = 2; id special_eos_id = 2;
id special_unk_id = 0; id special_unk_id = 0;
id special_sep_id = -1; id special_sep_id = LLAMA_TOKEN_NULL;
id special_pad_id = -1; id special_pad_id = LLAMA_TOKEN_NULL;
id special_cls_id = -1; id special_cls_id = LLAMA_TOKEN_NULL;
id special_mask_id = -1; id special_mask_id = LLAMA_TOKEN_NULL;
id linefeed_id = 13; id linefeed_id = 13;
id special_prefix_id = -1; id special_prefix_id = LLAMA_TOKEN_NULL;
id special_suffix_id = -1; id special_suffix_id = LLAMA_TOKEN_NULL;
id special_middle_id = -1; id special_middle_id = LLAMA_TOKEN_NULL;
id special_eot_id = -1; // TODO: move above after "eos_id", and here add "file separator" token id special_eot_id = LLAMA_TOKEN_NULL; // TODO: move above after "eos_id", and here add "file separator" token
id special_eom_id = -1; id special_eom_id = LLAMA_TOKEN_NULL;
// set of all tokens that cause "end of generation" // set of all tokens that cause "end of generation"
std::set<id> special_eog_ids; std::set<id> special_eog_ids;

2
src/llama.cpp
View File

@ -2412,7 +2412,7 @@ struct llama_hparams {
// needed by encoder-decoder models (e.g. T5, FLAN-T5) // needed by encoder-decoder models (e.g. T5, FLAN-T5)
// ref: https://github.com/ggerganov/llama.cpp/pull/8141 // ref: https://github.com/ggerganov/llama.cpp/pull/8141
llama_token dec_start_token_id = -1; llama_token dec_start_token_id = LLAMA_TOKEN_NULL;
enum llama_pooling_type pooling_type = LLAMA_POOLING_TYPE_NONE; enum llama_pooling_type pooling_type = LLAMA_POOLING_TYPE_NONE;
enum llama_rope_type rope_type = LLAMA_ROPE_TYPE_NONE; enum llama_rope_type rope_type = LLAMA_ROPE_TYPE_NONE;