feat: support image_input for image embeddings

README.md

@@ -123,6 +123,7 @@ You may use `/run` (asynchronous, start job and return job ID) or `/runsync` (sy
 Inputs:
 * `model`: name of one of the deployed models.
 * `input`: single text string or list of texts to embed
+* `image_input`: single url or list of urls to embed (for use with CLIP models)
 
 ### Reranking
 Inputs:

src/embedding_service.py

@@ -76,6 +76,28 @@ async def route_openai_get_embeddings(
                 embeddings, model=model_name, usage=usage
             )
 
+    async def route_get_image_embeddings(
+        self,
+        image_input: str | list[str],
+        model_name: str,
+        return_as_list: bool = False,
+    ):
+        """returns embeddings for the input image urls"""
+        if not self.is_running:
+            await self.start()
+        if not isinstance(image_input, list):
+            image_input = [image_input]
+
+        embeddings, usage = await self.engine_array[model_name].image_embed(image_input)
+        if return_as_list:
+            return [
+                list_embeddings_to_response(embeddings, model=model_name, usage=usage)
+            ]
+        else:
+            return list_embeddings_to_response(
+                embeddings, model=model_name, usage=usage
+            )
+
     async def infinity_rerank(
         self, query: str, docs: str, return_docs: str, model_name: str
     ):

src/handler.py

@@ -47,6 +47,12 @@ async def async_generator_handler(job: dict[str, Any]):
                 "embedding_input": job_input.get("input"),
                 "model_name": job_input.get("model"),
             }
+        # handle image urls (for image embeddings)
+        elif job_input.get("image_input"):
+            call_fn, kwargs = embedding_service.route_get_image_embeddings, {
+                "image_input": job_input.get("image_input"),
+                "model_name": job_input.get("model"),
+            }
         else:
             return create_error_response(f"Invalid input: {job}").model_dump()
     try: