tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-28 20:25:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : improve sep token handling (#14272)

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2025-06-20 14:04:09 +02:00
committed by GitHub
parent e28c1b93fd
commit 88fc854b4b
15 changed files with 161 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/llama-arch.cpp
View File

@ -198,6 +198,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
{ LLM_KV_TOKENIZER_MASK_ID, "tokenizer.ggml.mask_token_id" },
{ LLM_KV_TOKENIZER_ADD_BOS, "tokenizer.ggml.add_bos_token" },
{ LLM_KV_TOKENIZER_ADD_EOS, "tokenizer.ggml.add_eos_token" },
{ LLM_KV_TOKENIZER_ADD_SEP, "tokenizer.ggml.add_sep_token" },
{ LLM_KV_TOKENIZER_ADD_PREFIX, "tokenizer.ggml.add_space_prefix" },
{ LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, "tokenizer.ggml.remove_extra_whitespaces" },
{ LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" },

1
src/llama-arch.h
View File

@ -194,6 +194,7 @@ enum llm_kv {
LLM_KV_TOKENIZER_MASK_ID,
LLM_KV_TOKENIZER_ADD_BOS,
LLM_KV_TOKENIZER_ADD_EOS,
LLM_KV_TOKENIZER_ADD_SEP,
LLM_KV_TOKENIZER_ADD_PREFIX,
LLM_KV_TOKENIZER_REMOVE_EXTRA_WS,
LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP,

1
src/llama-model-saver.cpp
View File

@ -228,6 +228,7 @@ void llama_model_saver::add_kv_from_model() {
// add_kv(LLM_KV_TOKENIZER_MASK_ID, ???);
add_kv(LLM_KV_TOKENIZER_ADD_BOS, vocab.get_add_bos());
add_kv(LLM_KV_TOKENIZER_ADD_EOS, vocab.get_add_eos());
add_kv(LLM_KV_TOKENIZER_ADD_SEP, vocab.get_add_sep());
add_kv(LLM_KV_TOKENIZER_ADD_PREFIX, vocab.get_add_space_prefix());
add_kv(LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, vocab.get_remove_extra_whitespaces());
add_kv(LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, vocab.get_precompiled_charsmap());

24
src/llama-vocab.cpp
View File

@ -1269,6 +1269,7 @@ struct llama_vocab::impl {
bool add_space_prefix = false;
bool add_bos = false;
bool add_eos = false;
bool add_sep = false;
bool ignore_merges = false;
bool clean_spaces = false; // clean_up_tokenization_spaces
bool remove_extra_whitespaces = false;
@ -1421,6 +1422,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
special_sep_id = 102;
special_pad_id = 0;
special_mask_id = 103;
add_sep = true;
} else if (tokenizer_model == "gpt2") {
type = LLAMA_VOCAB_TYPE_BPE;
@ -1550,12 +1553,15 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
tokenizer_pre == "jina-es" ||
tokenizer_pre == "jina-de" ||
tokenizer_pre == "gigachat" ||
tokenizer_pre == "jina-v1-en" ||
tokenizer_pre == "jina-v2-es" ||
tokenizer_pre == "jina-v2-de" ||
tokenizer_pre == "jina-v2-de") {
pre_type = LLAMA_VOCAB_PRE_TYPE_GPT2;
} else if (
tokenizer_pre == "jina-v1-en" ||
tokenizer_pre == "jina-v2-code" ||
tokenizer_pre == "roberta-bpe") {
pre_type = LLAMA_VOCAB_PRE_TYPE_GPT2;
add_sep = true;
} else if (
tokenizer_pre == "refact") {
pre_type = LLAMA_VOCAB_PRE_TYPE_REFACT;
@ -1665,6 +1671,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
clean_spaces = true;
add_bos = true;
add_eos = false;
add_sep = true;
} else if (type == LLAMA_VOCAB_TYPE_UGM) {
pre_type = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
add_bos = false;
@ -1801,7 +1808,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
}
}
// Handle add_bos and add_eos
// Handle add_bos, add_eos and add_sep
{
bool temp = true;
@ -1811,6 +1818,9 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
if (ml.get_key(LLM_KV_TOKENIZER_ADD_EOS, temp, false)) {
add_eos = temp;
}
if (ml.get_key(LLM_KV_TOKENIZER_ADD_SEP, temp, false)) {
add_sep = temp;
}
}
// auto-detect special tokens by text
@ -3000,6 +3010,10 @@ bool llama_vocab::get_add_eos() const {
return pimpl->add_eos;
}
bool llama_vocab::get_add_sep() const {
return pimpl->add_sep;
}
bool llama_vocab::get_ignore_merges() const {
return pimpl->ignore_merges;
}
@ -3191,6 +3205,10 @@ bool llama_vocab_get_add_eos(const struct llama_vocab * vocab) {
return vocab->get_add_eos();
}
bool llama_vocab_get_add_sep(const struct llama_vocab * vocab) {
return vocab->get_add_sep();
}
llama_token llama_vocab_fim_pre(const struct llama_vocab * vocab) {
return vocab->token_fim_pre();
}

1
src/llama-vocab.h
View File

@ -74,6 +74,7 @@ struct llama_vocab {
bool get_add_space_prefix () const;
bool get_add_bos () const;
bool get_add_eos () const;
bool get_add_sep () const;
bool get_ignore_merges () const;
bool get_clean_spaces () const;
bool get_remove_extra_whitespaces () const;