Add boilerplate for vllm inference provider

Signed-off-by: Russell Bryant <rbryant@redhat.com>

llama_stack/providers/adapters/inference/__init__.py

@@ -1,5 +0,0 @@
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the terms described in the LICENSE file in
-# the root directory of this source tree.

llama_stack/providers/impls/vllm/__init__.py

@@ -0,0 +1,11 @@
+from typing import Any
+
+from .config import VLLMConfig
+
+
+async def get_provider_impl(config: VLLMConfig, _deps) -> Any:
+    from .vllm import VLLMInferenceImpl
+
+    impl = VLLMInferenceImpl(config)
+    await impl.initialize()
+    return impl

llama_stack/providers/impls/vllm/config.py

@@ -0,0 +1,5 @@
+from pydantic import BaseModel
+
+
+class VLLMConfig(BaseModel):
+    pass

llama_stack/providers/impls/vllm/vllm.py

@@ -0,0 +1,33 @@
+import logging
+from typing import Any
+
+from llama_stack.apis.inference.inference import CompletionResponse, CompletionResponseStreamChunk, LogProbConfig, ChatCompletionResponse, ChatCompletionResponseStreamChunk, EmbeddingsResponse
+from llama_stack.apis.inference import Inference
+
+from .config import VLLMConfig
+
+from llama_models.llama3.api.datatypes import InterleavedTextMedia, Message, ToolChoice, ToolDefinition, ToolPromptFormat
+
+
+log = logging.getLogger(__name__)
+
+
+class VLLMInferenceImpl(Inference):
+    """Inference implementation for vLLM."""
+    def __init__(self, config: VLLMConfig):
+        self.config = config
+
+    async def initialize(self):
+        log.info("Initializing vLLM inference adapter")
+        pass
+
+    async def completion(self, model: str, content: InterleavedTextMedia, sampling_params: Any | None = ..., stream: bool | None = False, logprobs: LogProbConfig | None = None) -> CompletionResponse | CompletionResponseStreamChunk:
+        log.info("vLLM completion")
+        return None
+
+    async def chat_completion(self, model: str, messages: list[Message], sampling_params: Any | None = ..., tools: list[ToolDefinition] | None = ..., tool_choice: ToolChoice | None = ..., tool_prompt_format: ToolPromptFormat | None = ..., stream: bool | None = False, logprobs: LogProbConfig | None = None) -> ChatCompletionResponse | ChatCompletionResponseStreamChunk:
+        log.info("vLLM chat completion")
+        return None
+
+    async def embeddings(self, model: str, contents: list[InterleavedTextMedia]) -> EmbeddingsResponse:
+        log.info("vLLM embeddings")