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

fix double bos issue (drop bos/eos tokens from jinja template)

Browse source
This commit is contained in:
Olivier Chafik 2025-02-03 13:59:16 +00:00
parent cf83623a47
commit 5d18d76b69
2 changed files with 13 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

10
common/common.cpp
View file

@ -1902,6 +1902,11 @@ common_chat_templates common_chat_templates_from_model(const struct llama_model
default_template_src = CHATML_TEMPLATE_SRC; default_template_src = CHATML_TEMPLATE_SRC;
} }
} }
std::string token_bos;
std::string token_eos;
// TODO: update logic that adds BOS and EOS tokens to the tokenized prompt, in favour of the template.
#if 0
auto vocab = llama_model_get_vocab(model);
const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) { const auto get_token = [&](llama_token token, const char * name, const char * jinja_variable_name) {
if (token == LLAMA_TOKEN_NULL) { if (token == LLAMA_TOKEN_NULL) {
if (default_template_src.find(jinja_variable_name) != std::string::npos if (default_template_src.find(jinja_variable_name) != std::string::npos
@ -1913,8 +1918,9 @@ common_chat_templates common_chat_templates_from_model(const struct llama_model
return common_token_to_piece(vocab, token, true); return common_token_to_piece(vocab, token, true);
} }
}; };
auto token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token"); token_bos = get_token(llama_vocab_bos(vocab), "BOS", "bos_token");
auto token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token"); token_eos = get_token(llama_vocab_eos(vocab), "EOS", "eos_token");
#endif
return { return {
has_explicit_template, has_explicit_template,
std::make_unique<minja::chat_template>(default_template_src, token_bos, token_eos), std::make_unique<minja::chat_template>(default_template_src, token_bos, token_eos),

3
examples/server/tests/unit/test_chat_completion.py
View file

@ -13,8 +13,11 @@ def create_server():
@pytest.mark.parametrize( @pytest.mark.parametrize(
"model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason,jinja,chat_template", "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason,jinja,chat_template",
[ [
(None, "Book", "Hey", 8, "But she couldn't", 69, 8, "length", False, None),
(None, "Book", "Hey", 8, "But she couldn't", 69, 8, "length", True, None),
(None, "Book", "What is the best book", 8, "(Suddenly)+|\\{ \" Sarax.", 77, 8, "length", False, None), (None, "Book", "What is the best book", 8, "(Suddenly)+|\\{ \" Sarax.", 77, 8, "length", False, None),
(None, "Book", "What is the best book", 8, "(Suddenly)+|\\{ \" Sarax.", 77, 8, "length", True, None), (None, "Book", "What is the best book", 8, "(Suddenly)+|\\{ \" Sarax.", 77, 8, "length", True, None),
(None, "Book", "What is the best book", 8, "(Suddenly)+|\\{ \" Sarax.", 77, 8, "length", True, 'chatml'),
(None, "Book", "What is the best book", 8, "^ blue", 23, 8, "length", True, "This is not a chat template, it is"), (None, "Book", "What is the best book", 8, "^ blue", 23, 8, "length", True, "This is not a chat template, it is"),
("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", False, None), ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", False, None),
("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", True, None), ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length", True, None),