remote::vllm now works with vision models

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -22,6 +22,9 @@ from llama_stack.providers.utils.inference.openai_compat import (
 )
 from llama_stack.providers.utils.inference.prompt_adapter import (
     chat_completion_request_to_prompt,
+    completion_request_to_prompt,
+    convert_message_to_dict,
+    request_has_media,
 )
 
 from .config import VLLMInferenceAdapterConfig
@@ -105,18 +108,24 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
     async def _nonstream_chat_completion(
         self, request: ChatCompletionRequest, client: OpenAI
     ) -> ChatCompletionResponse:
-        params = self._get_params(request)
+        params = await self._get_params(request)
+        if "messages" in params:
+            r = client.chat.completions.create(**params)
+        else:
             r = client.completions.create(**params)
         return process_chat_completion_response(r, self.formatter)
 
     async def _stream_chat_completion(
         self, request: ChatCompletionRequest, client: OpenAI
     ) -> AsyncGenerator:
-        params = self._get_params(request)
+        params = await self._get_params(request)
 
         # TODO: Can we use client.completions.acreate() or maybe there is another way to directly create an async
         #  generator so this wrapper is not necessary?
         async def _to_async_generator():
+            if "messages" in params:
+                s = client.chat.completions.create(**params)
+            else:
                 s = client.completions.create(**params)
             for chunk in s:
                 yield chunk
@@ -127,7 +136,9 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
         ):
             yield chunk
 
-    def _get_params(self, request: ChatCompletionRequest) -> dict:
+    async def _get_params(
+        self, request: Union[ChatCompletionRequest, CompletionRequest]
+    ) -> dict:
         options = get_sampling_options(request.sampling_params)
         if "max_tokens" not in options:
             options["max_tokens"] = self.config.max_tokens
@@ -136,9 +147,28 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
         if model is None:
             raise ValueError(f"Unknown model: {request.model}")
 
+        input_dict = {}
+        media_present = request_has_media(request)
+        if isinstance(request, ChatCompletionRequest):
+            if media_present:
+                # vllm does not seem to work well with image urls, so we download the images
+                input_dict["messages"] = [
+                    await convert_message_to_dict(m, download=True)
+                    for m in request.messages
+                ]
+            else:
+                input_dict["prompt"] = chat_completion_request_to_prompt(
+                    request, self.formatter
+                )
+        else:
+            assert (
+                not media_present
+            ), "Together does not support media for Completion requests"
+            input_dict["prompt"] = completion_request_to_prompt(request, self.formatter)
+
         return {
             "model": model.huggingface_repo,
-            "prompt": chat_completion_request_to_prompt(request, self.formatter),
+            **input_dict,
             "stream": request.stream,
             **options,
         }

llama_stack/providers/tests/inference/test_vision_inference.py

@@ -20,8 +20,25 @@ THIS_DIR = Path(__file__).parent
 
 class TestVisionModelInference:
     @pytest.mark.asyncio
+    @pytest.mark.parametrize(
+        "image, expected_strings",
+        [
+            (
+                ImageMedia(image=PIL_Image.open(THIS_DIR / "pasta.jpeg")),
+                ["spaghetti"],
+            ),
+            (
+                ImageMedia(
+                    image=URL(
+                        uri="https://www.healthypawspetinsurance.com/Images/V3/DogAndPuppyInsurance/Dog_CTA_Desktop_HeroImage.jpg"
+                    )
+                ),
+                ["puppy"],
+            ),
+        ],
+    )
     async def test_vision_chat_completion_non_streaming(
-        self, inference_model, inference_stack
+        self, inference_model, inference_stack, image, expected_strings
     ):
         inference_impl, _ = inference_stack
 
@@ -31,35 +48,20 @@ class TestVisionModelInference:
             "remote::together",
             "remote::fireworks",
             "remote::ollama",
+            "remote::vllm",
         ):
             pytest.skip(
                 "Other inference providers don't support vision chat completion() yet"
             )
 
-        images = [
-            ImageMedia(image=PIL_Image.open(THIS_DIR / "pasta.jpeg")),
-            ImageMedia(
-                image=URL(
-                    uri="https://www.healthypawspetinsurance.com/Images/V3/DogAndPuppyInsurance/Dog_CTA_Desktop_HeroImage.jpg"
-                )
-            ),
-        ]
-
-        # These are a bit hit-and-miss, need to be careful
-        expected_strings_to_check = [
-            ["spaghetti"],
-            ["puppy"],
-        ]
-        for image, expected_strings in zip(images, expected_strings_to_check):
         response = await inference_impl.chat_completion(
             model=inference_model,
             messages=[
-                    SystemMessage(content="You are a helpful assistant."),
-                    UserMessage(
-                        content=[image, "Describe this image in two sentences."]
-                    ),
+                UserMessage(content="You are a helpful assistant."),
+                UserMessage(content=[image, "Describe this image in two sentences."]),
             ],
             stream=False,
+            sampling_params=SamplingParams(max_tokens=100),
         )
 
         assert isinstance(response, ChatCompletionResponse)
@@ -80,6 +82,7 @@ class TestVisionModelInference:
             "remote::together",
             "remote::fireworks",
             "remote::ollama",
+            "remote::vllm",
         ):
             pytest.skip(
                 "Other inference providers don't support vision chat completion() yet"
@@ -101,12 +104,13 @@ class TestVisionModelInference:
                 async for r in await inference_impl.chat_completion(
                     model=inference_model,
                     messages=[
-                        SystemMessage(content="You are a helpful assistant."),
+                        UserMessage(content="You are a helpful assistant."),
                         UserMessage(
                             content=[image, "Describe this image in two sentences."]
                         ),
                     ],
                     stream=True,
+                    sampling_params=SamplingParams(max_tokens=100),
                 )
             ]
 

llama_stack/providers/utils/inference/prompt_adapter.py

@@ -90,13 +90,15 @@ async def convert_image_media_to_url(
         return base64.b64encode(content).decode("utf-8")
 
 
-async def convert_message_to_dict(message: Message) -> dict:
+# TODO: name this function better! this is about OpenAI compatibile image
+# media conversion of the message. this should probably go in openai_compat.py
+async def convert_message_to_dict(message: Message, download: bool = False) -> dict:
     async def _convert_content(content) -> dict:
         if isinstance(content, ImageMedia):
             return {
                 "type": "image_url",
                 "image_url": {
-                    "url": await convert_image_media_to_url(content),
+                    "url": await convert_image_media_to_url(content, download=download),
                 },
             }
         else: