[Anthropic][Frontend] Added chat_template_kwargs to /v1/messages (#40125)

Signed-off-by: Aleksandar Yanakiev <alexander.yanakiev@discretestack.com>
Co-authored-by: Aleksandar Yanakiev <alexander.yanakiev@discretestack.com>

vllm/entrypoints/anthropic/protocol.py

@@ -117,6 +117,13 @@ class AnthropicMessagesRequest(BaseModel):
         default=None,
         description="KVTransfer parameters used for disaggregated serving.",
     )
+    chat_template_kwargs: dict[str, Any] | None = Field(
+        default=None,
+        description=(
+            "Additional keyword args to pass to the chat template renderer. "
+            "Will be accessible by the template."
+        ),
+    )
 
     @field_validator("model")
     @classmethod
@@ -212,6 +219,15 @@ class AnthropicCountTokensRequest(BaseModel):
     tool_choice: AnthropicToolChoice | None = None
     tools: list[AnthropicTool] | None = None
 
+    # vLLM-specific fields that are not in Anthropic spec
+    chat_template_kwargs: dict[str, Any] | None = Field(
+        default=None,
+        description=(
+            "Additional keyword args to pass to the chat template renderer. "
+            "Will be accessible by the template."
+        ),
+    )
+
     @field_validator("model")
     @classmethod
     def validate_model(cls, v):

vllm/entrypoints/anthropic/serving.py

@@ -323,6 +323,7 @@ class AnthropicServingMessages(OpenAIServingChat):
             return ChatCompletionRequest(
                 model=anthropic_request.model,
                 messages=openai_messages,
+                chat_template_kwargs=anthropic_request.chat_template_kwargs,
             )
 
         return ChatCompletionRequest(
@@ -335,6 +336,7 @@ class AnthropicServingMessages(OpenAIServingChat):
             top_p=anthropic_request.top_p,
             top_k=anthropic_request.top_k,
             kv_transfer_params=anthropic_request.kv_transfer_params,
+            chat_template_kwargs=anthropic_request.chat_template_kwargs,
         )
 
     @classmethod