fix: Forward stream_options through multimodal request pipeline (#6474)

Signed-off-by: Blake Ledden <blake.ledden@gmail.com>

components/src/dynamo/vllm/multimodal_utils/protocol.py

@@ -157,6 +157,7 @@ class MultiModalRequest(BaseModel):
     max_tokens: Optional[int] = None
     temperature: Optional[float] = None
     stream: Optional[bool] = True
+    stream_options: Optional[dict] = None
 
 
 class MultiModalInput(BaseModel):

examples/multimodal/components/processor.py

@@ -221,6 +221,7 @@ class Processor(ProcessMixIn):
             model=raw_request.model,
             messages=[msg],
             stream=True,
+            stream_options=raw_request.stream_options,
             max_tokens=raw_request.max_tokens,
             temperature=raw_request.temperature,
             request_id=str(uuid.uuid4()),

examples/multimodal/utils/protocol.py

@@ -146,6 +146,7 @@ class MultiModalRequest(BaseModel):
     max_tokens: Optional[int] = None
     temperature: Optional[float] = None
     stream: Optional[bool] = True
+    stream_options: Optional[dict] = None
 
 
 class MultiModalInput(BaseModel):