xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-08 11:55:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

treat do_lower_case in the same way as the sentence-transformers library (#11815)

Browse Source 
Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
This commit is contained in:
Maximilien de Bayser 2025-01-09 00:05:43 -03:00 committed by GitHub
parent 615e4a5401
commit 1fe554bac3
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
5 changed files with 13 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
tests/entrypoints/openai/test_serving_chat.py
View File

@ -35,6 +35,7 @@ class MockModelConfig:
logits_processor_pattern = None logits_processor_pattern = None
diff_sampling_param: Optional[dict] = None diff_sampling_param: Optional[dict] = None
allowed_local_media_path: str = "" allowed_local_media_path: str = ""
encoder_config = None
def get_diff_sampling_param(self): def get_diff_sampling_param(self):
return self.diff_sampling_param or {} return self.diff_sampling_param or {}

1
tests/models/embedding/language/test_embedding.py
View File

@ -15,6 +15,7 @@ from ..utils import check_embeddings_close
# [Encoder-only] # [Encoder-only]
pytest.param("BAAI/bge-base-en-v1.5", pytest.param("BAAI/bge-base-en-v1.5",
marks=[pytest.mark.core_model, pytest.mark.cpu_model]), marks=[pytest.mark.core_model, pytest.mark.cpu_model]),
pytest.param("sentence-transformers/all-MiniLM-L12-v2"),
pytest.param("intfloat/multilingual-e5-large"), pytest.param("intfloat/multilingual-e5-large"),
# [Encoder-decoder] # [Encoder-decoder]
pytest.param("intfloat/e5-mistral-7b-instruct", pytest.param("intfloat/e5-mistral-7b-instruct",

5
vllm/entrypoints/openai/serving_engine.py
View File

@ -160,6 +160,11 @@ class OpenAIServing:
truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]], truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]],
add_special_tokens: bool, add_special_tokens: bool,
) -> TextTokensPrompt: ) -> TextTokensPrompt:
if (self.model_config.encoder_config is not None
and self.model_config.encoder_config.get(
"do_lower_case", False)):
prompt = prompt.lower()
if truncate_prompt_tokens is None: if truncate_prompt_tokens is None:
encoded = tokenizer(prompt, add_special_tokens=add_special_tokens) encoded = tokenizer(prompt, add_special_tokens=add_special_tokens)
else: else:

6
vllm/inputs/preprocess.py
View File

@ -190,6 +190,12 @@ class InputPreprocessor:
# on the task and language of their request. Also needed to avoid # on the task and language of their request. Also needed to avoid
# appending an EOS token to the prompt which disrupts generation. # appending an EOS token to the prompt which disrupts generation.
add_special_tokens = False add_special_tokens = False
if (self.model_config.encoder_config is not None
and self.model_config.encoder_config.get(
"do_lower_case", False)):
prompt = prompt.lower()
return tokenizer.encode(request_id=request_id, return tokenizer.encode(request_id=request_id,
prompt=prompt, prompt=prompt,
lora_request=lora_request, lora_request=lora_request,

5
vllm/transformers_utils/tokenizer_group/__init__.py
View File

@ -26,11 +26,6 @@ def init_tokenizer_from_configs(model_config: ModelConfig,
trust_remote_code=model_config.trust_remote_code, trust_remote_code=model_config.trust_remote_code,
revision=model_config.tokenizer_revision) revision=model_config.tokenizer_revision)
if (model_config.encoder_config is not None
and "do_lower_case" in model_config.encoder_config):
init_kwargs["do_lower_case"] = model_config.encoder_config[
"do_lower_case"]
return get_tokenizer_group(parallel_config.tokenizer_pool_config, return get_tokenizer_group(parallel_config.tokenizer_pool_config,
**init_kwargs) **init_kwargs)