refactor(passthrough): use AsyncOpenAI instead of AsyncLlamaStackClient (#4085)

We'd like to remove the dependence of `llama-stack` on
`llama-stack-client`. This is a necessary step.

A few small cleanups
- Enables `embeddings` now also
- Remove ModelRegistryHelper dependency (unused)
- Consolidate to auth_credential field via RemoteInferenceProviderConfig
- Implement list_models() to fetch from downstream /v1/models

## Test Plan

Tested using this script
https://gist.github.com/ashwinb/6356463d10f989c0682ab3bff8589581

Output:
```
Listing models from downstream server...
Available models: ['passthrough/ollama/nomic-embed-text:latest', 'passthrough/ollama/all-minilm:l6-v2', 'passthrough/ollama/llama3.2-vision:11b', 'passthrough/ollama/llama3.2-vision:latest', 'passthrough/ollama/llama-guard3:1b', 'passthrough/o
llama/llama3.2:1b', 'passthrough/ollama/all-minilm:latest', 'passthrough/ollama/llama3.2:3b', 'passthrough/ollama/llama3.2:3b-instruct-fp16', 'passthrough/bedrock/meta.llama3-1-8b-instruct-v1:0', 'passthrough/bedrock/meta.llama3-1-70b-instruct
-v1:0', 'passthrough/bedrock/meta.llama3-1-405b-instruct-v1:0', 'passthrough/sentence-transformers/nomic-ai/nomic-embed-text-v1.5']

Using LLM model: passthrough/ollama/llama3.2-vision:11b

Making inference request...

Response: 4.

--- Testing streaming ---
Streamed response: ChatCompletionChunk(id='chatcmpl-64', choices=[Choice(delta=ChoiceDelta(content='1', reasoning_content=None, refusal=None, role='assistant', tool_calls=None), finish_reason='', index=0, logprobs=None)], created=1762381674, m
odel='passthrough/ollama/llama3.2-vision:11b', object='chat.completion.chunk', usage=None)
...
5ChatCompletionChunk(id='chatcmpl-64', choices=[Choice(delta=ChoiceDelta(content='', reasoning_content=None, refusal=None, role='assistant', tool_calls=None), finish_reason='stop', index=0, logprobs=None)], created=1762381674, model='passthrou
gh/ollama/llama3.2-vision:11b', object='chat.completion.chunk', usage=None)
```

docs/docs/providers/inference/remote_passthrough.mdx

@@ -16,7 +16,7 @@ Passthrough inference provider for connecting to any external inference service
 |-------|------|----------|---------|-------------|
 | `allowed_models` | `list[str \| None` | No |  | List of models that should be registered with the model registry. If None, all models are allowed. |
 | `refresh_models` | `<class 'bool'>` | No | False | Whether to refresh models periodically from the provider |
-| `api_key` | `pydantic.types.SecretStr \| None` | No |  | API Key for the passthrouth endpoint |
+| `api_key` | `pydantic.types.SecretStr \| None` | No |  | Authentication credential for the provider |
 | `url` | `<class 'str'>` | No |  | The URL for the passthrough endpoint |
 
 ## Sample Configuration

src/llama_stack/providers/remote/inference/passthrough/__init__.py

@@ -10,8 +10,8 @@ from .config import PassthroughImplConfig
 
 
 class PassthroughProviderDataValidator(BaseModel):
-    url: str
+    passthrough_url: str
-    api_key: str
+    passthrough_api_key: str
 
 
 async def get_adapter_impl(config: PassthroughImplConfig, _deps):

src/llama_stack/providers/remote/inference/passthrough/config.py

@@ -6,7 +6,7 @@
 
 from typing import Any
 
-from pydantic import Field, SecretStr
+from pydantic import Field
 
 from llama_stack.providers.utils.inference.model_registry import RemoteInferenceProviderConfig
 from llama_stack.schema_utils import json_schema_type
@@ -19,11 +19,6 @@ class PassthroughImplConfig(RemoteInferenceProviderConfig):
         description="The URL for the passthrough endpoint",
     )
 
-    api_key: SecretStr | None = Field(
-        default=None,
-        description="API Key for the passthrouth endpoint",
-    )
-
     @classmethod
     def sample_run_config(
         cls, url: str = "${env.PASSTHROUGH_URL}", api_key: str = "${env.PASSTHROUGH_API_KEY}", **kwargs

src/llama_stack/providers/remote/inference/passthrough/passthrough.py

@@ -5,9 +5,8 @@
 # the root directory of this source tree.
 
 from collections.abc import AsyncIterator
-from typing import Any
 
-from llama_stack_client import AsyncLlamaStackClient
+from openai import AsyncOpenAI
 
 from llama_stack.apis.inference import (
     Inference,
@@ -20,103 +19,117 @@ from llama_stack.apis.inference import (
     OpenAIEmbeddingsResponse,
 )
 from llama_stack.apis.models import Model
-from llama_stack.core.library_client import convert_pydantic_to_json_value
+from llama_stack.core.request_headers import NeedsRequestProviderData
-from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
 
 from .config import PassthroughImplConfig
 
 
-class PassthroughInferenceAdapter(Inference):
+class PassthroughInferenceAdapter(NeedsRequestProviderData, Inference):
     def __init__(self, config: PassthroughImplConfig) -> None:
-        ModelRegistryHelper.__init__(self)
         self.config = config
 
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        pass
+
     async def unregister_model(self, model_id: str) -> None:
         pass
 
     async def register_model(self, model: Model) -> Model:
         return model
 
-    def _get_client(self) -> AsyncLlamaStackClient:
+    async def list_models(self) -> list[Model]:
-        passthrough_url = None
+        """List models by calling the downstream /v1/models endpoint."""
-        passthrough_api_key = None
+        client = self._get_openai_client()
-        provider_data = None
 
-        if self.config.url is not None:
+        response = await client.models.list()
-            passthrough_url = self.config.url
-        else:
-            provider_data = self.get_request_provider_data()
-            if provider_data is None or not provider_data.passthrough_url:
-                raise ValueError(
-                    'Pass url of the passthrough endpoint in the header X-LlamaStack-Provider-Data as { "passthrough_url": <your passthrough url>}'
-                )
-            passthrough_url = provider_data.passthrough_url
 
-        if self.config.api_key is not None:
+        # Convert from OpenAI format to Llama Stack Model format
-            passthrough_api_key = self.config.api_key.get_secret_value()
+        models = []
-        else:
+        for model_data in response.data:
-            provider_data = self.get_request_provider_data()
+            downstream_model_id = model_data.id
-            if provider_data is None or not provider_data.passthrough_api_key:
+            custom_metadata = getattr(model_data, "custom_metadata", {}) or {}
-                raise ValueError(
-                    'Pass API Key for the passthrough endpoint in the header X-LlamaStack-Provider-Data as { "passthrough_api_key": <your api key>}'
-                )
-            passthrough_api_key = provider_data.passthrough_api_key
 
-        return AsyncLlamaStackClient(
+            # Prefix identifier with provider ID for local registry
-            base_url=passthrough_url,
+            local_identifier = f"{self.__provider_id__}/{downstream_model_id}"
-            api_key=passthrough_api_key,
+
-            provider_data=provider_data,
+            model = Model(
+                identifier=local_identifier,
+                provider_id=self.__provider_id__,
+                provider_resource_id=downstream_model_id,
+                model_type=custom_metadata.get("model_type", "llm"),
+                metadata=custom_metadata,
+            )
+            models.append(model)
+
+        return models
+
+    async def should_refresh_models(self) -> bool:
+        """Passthrough should refresh models since they come from downstream dynamically."""
+        return self.config.refresh_models
+
+    def _get_openai_client(self) -> AsyncOpenAI:
+        """Get an AsyncOpenAI client configured for the downstream server."""
+        base_url = self._get_passthrough_url()
+        api_key = self._get_passthrough_api_key()
+
+        return AsyncOpenAI(
+            base_url=f"{base_url.rstrip('/')}/v1",
+            api_key=api_key,
         )
 
-    async def openai_embeddings(
+    def _get_passthrough_url(self) -> str:
-        self,
+        """Get the passthrough URL from config or provider data."""
-        params: OpenAIEmbeddingsRequestWithExtraBody,
+        if self.config.url is not None:
-    ) -> OpenAIEmbeddingsResponse:
+            return self.config.url
-        raise NotImplementedError()
+
+        provider_data = self.get_request_provider_data()
+        if provider_data is None:
+            raise ValueError(
+                'Pass url of the passthrough endpoint in the header X-LlamaStack-Provider-Data as { "passthrough_url": <your passthrough url>}'
+            )
+        return provider_data.passthrough_url
+
+    def _get_passthrough_api_key(self) -> str:
+        """Get the passthrough API key from config or provider data."""
+        if self.config.auth_credential is not None:
+            return self.config.auth_credential.get_secret_value()
+
+        provider_data = self.get_request_provider_data()
+        if provider_data is None:
+            raise ValueError(
+                'Pass API Key for the passthrough endpoint in the header X-LlamaStack-Provider-Data as { "passthrough_api_key": <your api key>}'
+            )
+        return provider_data.passthrough_api_key
 
     async def openai_completion(
         self,
         params: OpenAICompletionRequestWithExtraBody,
     ) -> OpenAICompletion:
-        client = self._get_client()
+        """Forward completion request to downstream using OpenAI client."""
-        model_obj = await self.model_store.get_model(params.model)
+        client = self._get_openai_client()
-
-        params = params.model_copy()
-        params.model = model_obj.provider_resource_id
-
         request_params = params.model_dump(exclude_none=True)
-
+        response = await client.completions.create(**request_params)
-        return await client.inference.openai_completion(**request_params)
+        return response  # type: ignore
 
     async def openai_chat_completion(
         self,
         params: OpenAIChatCompletionRequestWithExtraBody,
     ) -> OpenAIChatCompletion | AsyncIterator[OpenAIChatCompletionChunk]:
-        client = self._get_client()
+        """Forward chat completion request to downstream using OpenAI client."""
-        model_obj = await self.model_store.get_model(params.model)
+        client = self._get_openai_client()
-
-        params = params.model_copy()
-        params.model = model_obj.provider_resource_id
-
         request_params = params.model_dump(exclude_none=True)
+        response = await client.chat.completions.create(**request_params)
+        return response  # type: ignore
 
-        return await client.inference.openai_chat_completion(**request_params)
+    async def openai_embeddings(
-
+        self,
-    def cast_value_to_json_dict(self, request_params: dict[str, Any]) -> dict[str, Any]:
+        params: OpenAIEmbeddingsRequestWithExtraBody,
-        json_params = {}
+    ) -> OpenAIEmbeddingsResponse:
-        for key, value in request_params.items():
+        """Forward embeddings request to downstream using OpenAI client."""
-            json_input = convert_pydantic_to_json_value(value)
+        client = self._get_openai_client()
-            if isinstance(json_input, dict):
+        request_params = params.model_dump(exclude_none=True)
-                json_input = {k: v for k, v in json_input.items() if v is not None}
+        response = await client.embeddings.create(**request_params)
-            elif isinstance(json_input, list):
+        return response  # type: ignore
-                json_input = [x for x in json_input if x is not None]
-                new_input = []
-                for x in json_input:
-                    if isinstance(x, dict):
-                        x = {k: v for k, v in x.items() if v is not None}
-                    new_input.append(x)
-                json_input = new_input
-
-            json_params[key] = json_input
-
-        return json_params