fix: make vision and embedding tests pass with openai, anthropic and gemini

NOTE - Anthropic embeddings do not work due to LiteLLM not supporting
them.

llama_stack/providers/remote/inference/openai/models.py

@@ -20,11 +20,11 @@ MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
     ProviderModelEntry(
         provider_model_id="openai/text-embedding-3-small",
         model_type=ModelType.embedding,
-        metadata={"embedding_dimension": 1536},
+        metadata={"embedding_dimension": 1536, "context_length": 8192},
     ),
     ProviderModelEntry(
         provider_model_id="openai/text-embedding-3-large",
         model_type=ModelType.embedding,
-        metadata={"embedding_dimension": 3072},
+        metadata={"embedding_dimension": 3072, "context_length": 8192},
     ),
 ]

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -97,7 +97,6 @@ class LiteLLMOpenAIMixin(
         )
 
         params = await self._get_params(request)
-
         # unfortunately, we need to use synchronous litellm.completion here because litellm
         # caches various httpx.client objects in a non-eventloop aware manner
         response = litellm.completion(**params)

llama_stack/providers/utils/inference/openai_compat.py

@@ -523,15 +523,19 @@ async def convert_message_to_openai_dict_new(message: Message | Dict) -> OpenAIC
     ) -> Union[str, Iterable[OpenAIChatCompletionContentPartParam]]:
         # Llama Stack and OpenAI spec match for str and text input
         if isinstance(content, str):
-            return content
+            return OpenAIChatCompletionContentPartTextParam(
+                type="text",
+                text=content,
+            )
         elif isinstance(content, TextContentItem):
             return OpenAIChatCompletionContentPartTextParam(
+                type="text",
                 text=content.text,
             )
         elif isinstance(content, ImageContentItem):
             return OpenAIChatCompletionContentPartImageParam(
-                image_url=OpenAIImageURL(url=await convert_image_content_to_url(content)),
                 type="image_url",
+                image_url=OpenAIImageURL(url=await convert_image_content_to_url(content)),
             )
         elif isinstance(content, List):
             return [await _convert_user_message_content(item) for item in content]

llama_stack/templates/dev/run.yaml

@@ -119,12 +119,14 @@ models:
   model_type: llm
 - metadata:
     embedding_dimension: 1536
+    context_length: 8192
   model_id: openai/text-embedding-3-small
   provider_id: openai
   provider_model_id: openai/text-embedding-3-small
   model_type: embedding
 - metadata:
     embedding_dimension: 3072
+    context_length: 8192
   model_id: openai/text-embedding-3-large
   provider_id: openai
   provider_model_id: openai/text-embedding-3-large