feat(utils): add vision model type

Signed-off-by: Max Wittig <[email protected]>

Author: max-wittig

src/vllm_router/utils.py

@@ -66,12 +66,24 @@ def __call__(cls, *args, **kwargs):
 
 
 class ModelType(enum.Enum):
-    chat = "/v1/chat/completions"
-    completion = "/v1/completions"
-    embeddings = "/v1/embeddings"
-    rerank = "/v1/rerank"
-    score = "/v1/score"
-    transcription = "/v1/audio/transcriptions"
+    chat = "chat"
+    completion = "completion"
+    embeddings = "embeddings"
+    rerank = "rerank"
+    score = "score"
+    transcription = "transcription"
+    vision = "vision"
+
+    @staticmethod
+    def get_url(model_type: str):
+        match ModelType[model_type]:
+            case ModelType.chat: return "/v1/chat/completions"
+            case ModelType.completion: return "/v1/completions"
+            case ModelType.embeddings: return "/v1/embeddings"
+            case ModelType.rerank: return "/v1/rerank"
+            case ModelType.score: return "/v1/score"
+            case ModelType.transcription: return "/v1/audio/transcriptions"
+            case ModelType.vision: return "/v1/chat/completions"
 
     @staticmethod
     def get_test_payload(model_type: str):
@@ -101,6 +113,26 @@ def get_test_payload(model_type: str):
                     return {
                         "file": ("empty.wav", _SILENT_WAV_BYTES, "audio/wav"),
                     }
+            case ModelType.vision:
+                return {
+                    "messages": [
+                        {
+                            "role": "user",
+                            "content": [
+                                {
+                                    "type": "text",
+                                    "text": "This is a test. Just reply with yes",
+                                },
+                                {
+                                    "type": "image_url",
+                                    "image_url": {
+                                        "url": "data:image/jpeg;base64,iVBORw0KGgoAAAANSUhEUgAAAAIAAAACCAIAAAD91JpzAAAAG0lEQVR4nGLinfJq851wJn69udZSvIAAAAD//yf3BLKCfW8HAAAAAElFTkSuQmCC"
+                                    },
+                                },
+                            ],
+                        }
+                    ]
+                }
 
     @staticmethod
     def get_all_fields():
@@ -186,27 +218,24 @@ def update_content_length(request: Request, request_body: str):
 
 
 def is_model_healthy(url: str, model: str, model_type: str) -> bool:
-    model_details = ModelType[model_type]
+    model_url = ModelType.get_url(model_type)
 
     try:
         if model_type == "transcription":
-
             # for transcription, the backend expects multipart/form-data with a file
             # we will use pre-generated silent wav bytes
-            files = {"file": ("empty.wav", _SILENT_WAV_BYTES, "audio/wav")}
-            data = {"model": model}
             response = requests.post(
-                f"{url}{model_details.value}",
-                files=files,  # multipart/form-data
-                data=data,
+                f"{url}{model_url}",
+                files=ModelType[model_type].get_test_payload(model_type),  # multipart/form-data
+                data={"model": model},
                 timeout=10,
             )
         else:
             # for other model types (chat, completion, etc.)
             response = requests.post(
-                f"{url}{model_details.value}",
+                f"{url}{model_url}",
                 headers={"Content-Type": "application/json"},
-                json={"model": model} | model_details.get_test_payload(model_type),
+                json={"model": model} | ModelType[model_type].get_test_payload(model_type),
                 timeout=10,
             )