vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : use LLAMA_TOKEN_NULL (#11062)

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2025-01-06 10:52:15 +02:00 • committed by GitHub
parent 5047dd3546
commit 727368c60f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
11 changed files with 53 additions and 54 deletions
Download patch file Download diff file Expand all files Collapse all files

24
src/llama-vocab.cpp
View file

@ -497,7 +497,7 @@ struct llm_tokenizer_bpe_session {
bool append_bos(std::vector<llama_vocab::id> & output) const {
if (vocab.tokenizer_add_bos) {
GGML_ASSERT(vocab.special_bos_id != -1);
GGML_ASSERT(vocab.special_bos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_bos_id);
return true;
}
@ -506,7 +506,7 @@ struct llm_tokenizer_bpe_session {
bool append_eos(std::vector<llama_vocab::id> & output) const {
if (vocab.tokenizer_add_eos) {
GGML_ASSERT(vocab.special_eos_id != -1);
GGML_ASSERT(vocab.special_eos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_eos_id);
return true;
}
@ -1403,7 +1403,7 @@ static void tokenizer_st_partition(const llama_vocab & vocab, std::forward_list<
if (source == 0) {
buffer.erase_after(buffer.before_begin());
} else {
buffer.erase_after(std::next(buffer.begin(), (source-1)));
buffer.erase_after(std::next(buffer.begin(), (source - 1)));
}
// repeat for the right side
@ -1417,7 +1417,7 @@ static void tokenizer_st_partition(const llama_vocab & vocab, std::forward_list<
if (source == 0) {
buffer.erase_after(buffer.before_begin());
} else {
buffer.erase_after(std::next(buffer.begin(), (source-1)));
buffer.erase_after(std::next(buffer.begin(), (source - 1)));
}
break;
}
@ -1454,7 +1454,7 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
bool is_prev_special = true; // prefix with space if first token
if (add_special && vocab.tokenizer_add_bos) {
GGML_ASSERT(vocab.special_bos_id != -1);
GGML_ASSERT(vocab.special_bos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_bos_id);
is_prev_special = true;
}
@ -1489,7 +1489,7 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
}
if (add_special && vocab.tokenizer_add_eos) {
GGML_ASSERT(vocab.special_eos_id != -1);
GGML_ASSERT(vocab.special_eos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_eos_id);
}
} break;
@ -1522,7 +1522,7 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
case LLAMA_VOCAB_TYPE_WPM:
{
if (add_special) {
GGML_ASSERT(vocab.special_cls_id != -1);
GGML_ASSERT(vocab.special_cls_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_cls_id);
}
@ -1542,14 +1542,14 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
}
if (add_special) {
GGML_ASSERT(vocab.special_sep_id != -1);
GGML_ASSERT(vocab.special_sep_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_sep_id);
}
} break;
case LLAMA_VOCAB_TYPE_UGM:
{
if (add_special && vocab.tokenizer_add_bos) {
GGML_ASSERT(vocab.special_bos_id != -1);
GGML_ASSERT(vocab.special_bos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_bos_id);
}
llm_tokenizer_ugm_session session(vocab);
@ -1574,7 +1574,7 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
}
if (add_special && vocab.tokenizer_add_eos) {
GGML_ASSERT(vocab.special_eos_id != -1);
GGML_ASSERT(vocab.special_eos_id != LLAMA_TOKEN_NULL);
output.push_back(vocab.special_eos_id);
}
} break;
@ -1642,7 +1642,7 @@ llama_token_attr llama_token_get_attr_impl(const struct llama_vocab & vocab, lla
}
bool llama_token_is_eog_impl(const struct llama_vocab & vocab, llama_token token) {
return token != -1 && vocab.special_eog_ids.count(token) > 0;
return token != LLAMA_TOKEN_NULL && vocab.special_eog_ids.count(token) > 0;
}
bool llama_token_is_control_impl(const struct llama_vocab & vocab, llama_token token) {
@ -1881,7 +1881,7 @@ int32_t llama_detokenize_impl(
}
if (remove_special && vocab.tokenizer_add_eos) {
if (n_tokens > 0 && tokens[n_tokens-1] == vocab.special_eos_id) {
if (n_tokens > 0 && tokens[n_tokens - 1] == vocab.special_eos_id) {
n_tokens--;
}
}