Support add_generation_prompt in embeddings endpoint with chat request (#23931)

Signed-off-by: biba10 <[email protected]>

Author: biba10

vllm/entrypoints/openai/protocol.py

@@ -1342,6 +1342,14 @@ class EmbeddingChatRequest(OpenAIBaseModel):
     truncate_prompt_tokens: Optional[Annotated[int, Field(ge=-1)]] = None
 
     # --8<-- [start:chat-embedding-extra-params]
+    add_generation_prompt: bool = Field(
+        default=False,
+        description=
+        ("If true, the generation prompt will be added to the chat template. "
+         "This is a parameter used by chat template in tokenizer config of the "
+         "model."),
+    )
+
     add_special_tokens: bool = Field(
         default=False,
         description=(

vllm/entrypoints/openai/serving_embedding.py

@@ -93,9 +93,7 @@ async def _preprocess(
                     or ctx.chat_template,
                     chat_template_content_format=ctx.
                     chat_template_content_format,
-                    # In embedding requests, we are not generating tokens,
-                    # so there is no need to append extra tokens to the input
-                    add_generation_prompt=False,
+                    add_generation_prompt=ctx.request.add_generation_prompt,
                     continue_final_message=False,
                     add_special_tokens=ctx.request.add_special_tokens,
                 )