treat do_lower_case in the same way as the sentence-transformers library

tests/models/embedding/language/test_embedding.py

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

vllm/entrypoints/openai/serving_engine.py

@@ -160,6 +160,11 @@ def _normalize_prompt_text_to_input(
         truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]],
         add_special_tokens: bool,
     ) -> 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:
             encoded = tokenizer(prompt, add_special_tokens=add_special_tokens)
         else:

vllm/inputs/preprocess.py

@@ -190,6 +190,12 @@ def _tokenize_prompt(
             # on the task and language of their request. Also needed to avoid
             # appending an EOS token to the prompt which disrupts generation.
             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,
                                 prompt=prompt,
                                 lora_request=lora_request,

vllm/transformers_utils/tokenizer_group/__init__.py

@@ -26,11 +26,6 @@ def init_tokenizer_from_configs(model_config: ModelConfig,
                        trust_remote_code=model_config.trust_remote_code,
                        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,
                                **init_kwargs)