[#432] Add Groq Provider - tool calls

README.md

@@ -84,6 +84,7 @@ Additionally, we have designed every element of the Stack such that APIs as well
 |                                         Fireworks                                          |         Hosted         | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: |                    |                    |
 |                                        AWS Bedrock                                         |         Hosted         |                    | :heavy_check_mark: |                    | :heavy_check_mark: |                    |
 |                                          Together                                          |         Hosted         | :heavy_check_mark: | :heavy_check_mark: |                    | :heavy_check_mark: |                    |
+|                                            Groq                                            |         Hosted         |                    | :heavy_check_mark: |                    |                    |                    |
 |                                           Ollama                                           |      Single Node       |                    | :heavy_check_mark: |                    |                    |                    |
 |                                            TGI                                             | Hosted and Single Node |                    | :heavy_check_mark: |                    |                    |                    |
 | [NVIDIA NIM](https://build.nvidia.com/nim?filters=nimType%3Anim_type_run_anywhere&q=llama) | Hosted and Single Node |                    | :heavy_check_mark: |                    |                    |                    |

llama_stack/providers/registry/inference.py

@@ -149,6 +149,16 @@ def available_providers() -> List[ProviderSpec]:
                 provider_data_validator="llama_stack.providers.remote.inference.together.TogetherProviderDataValidator",
             ),
         ),
+        remote_provider_spec(
+            api=Api.inference,
+            adapter=AdapterSpec(
+                adapter_type="groq",
+                pip_packages=["groq"],
+                module="llama_stack.providers.remote.inference.groq",
+                config_class="llama_stack.providers.remote.inference.groq.GroqConfig",
+                provider_data_validator="llama_stack.providers.remote.inference.groq.GroqProviderDataValidator",
+            ),
+        ),
         remote_provider_spec(
             api=Api.inference,
             adapter=AdapterSpec(

llama_stack/providers/remote/inference/groq/__init__.py

@@ -0,0 +1,26 @@
+# 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.
+
+from pydantic import BaseModel
+
+from llama_stack.apis.inference import Inference
+
+from .config import GroqConfig
+
+
+class GroqProviderDataValidator(BaseModel):
+    groq_api_key: str
+
+
+async def get_adapter_impl(config: GroqConfig, _deps) -> Inference:
+    # import dynamically so the import is used only when it is needed
+    from .groq import GroqInferenceAdapter
+
+    if not isinstance(config, GroqConfig):
+        raise RuntimeError(f"Unexpected config type: {type(config)}")
+
+    adapter = GroqInferenceAdapter(config)
+    return adapter

llama_stack/providers/remote/inference/groq/config.py

@@ -0,0 +1,19 @@
+# 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.
+
+from typing import Optional
+
+from llama_models.schema_utils import json_schema_type
+from pydantic import BaseModel, Field
+
+
+@json_schema_type
+class GroqConfig(BaseModel):
+    api_key: Optional[str] = Field(
+        # The Groq client library loads the GROQ_API_KEY environment variable by default
+        default=None,
+        description="The Groq API key",
+    )

llama_stack/providers/remote/inference/groq/groq.py

@@ -0,0 +1,160 @@
+# 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.
+
+import warnings
+from typing import AsyncIterator, List, Optional, Union
+
+import groq
+from groq import Groq
+from llama_models.datatypes import SamplingParams
+from llama_models.llama3.api.datatypes import ToolDefinition, ToolPromptFormat
+from llama_models.sku_list import CoreModelId
+
+from llama_stack.apis.inference import (
+    ChatCompletionRequest,
+    ChatCompletionResponse,
+    ChatCompletionResponseStreamChunk,
+    CompletionResponse,
+    CompletionResponseStreamChunk,
+    EmbeddingsResponse,
+    Inference,
+    InterleavedContent,
+    LogProbConfig,
+    Message,
+    ResponseFormat,
+    ToolChoice,
+)
+from llama_stack.distribution.request_headers import NeedsRequestProviderData
+from llama_stack.providers.remote.inference.groq.config import GroqConfig
+from llama_stack.providers.utils.inference.model_registry import (
+    build_model_alias,
+    build_model_alias_with_just_provider_model_id,
+    ModelRegistryHelper,
+)
+from .groq_utils import (
+    convert_chat_completion_request,
+    convert_chat_completion_response,
+    convert_chat_completion_response_stream,
+)
+
+_MODEL_ALIASES = [
+    build_model_alias(
+        "llama3-8b-8192",
+        CoreModelId.llama3_1_8b_instruct.value,
+    ),
+    build_model_alias_with_just_provider_model_id(
+        "llama-3.1-8b-instant",
+        CoreModelId.llama3_1_8b_instruct.value,
+    ),
+    build_model_alias(
+        "llama3-70b-8192",
+        CoreModelId.llama3_70b_instruct.value,
+    ),
+    build_model_alias(
+        "llama-3.3-70b-versatile",
+        CoreModelId.llama3_3_70b_instruct.value,
+    ),
+    # Groq only contains a preview version for llama-3.2-3b
+    # Preview models aren't recommended for production use, but we include this one
+    # to pass the test fixture
+    # TODO(aidand): Replace this with a stable model once Groq supports it
+    build_model_alias(
+        "llama-3.2-3b-preview",
+        CoreModelId.llama3_2_3b_instruct.value,
+    ),
+]
+
+
+class GroqInferenceAdapter(Inference, ModelRegistryHelper, NeedsRequestProviderData):
+    _config: GroqConfig
+
+    def __init__(self, config: GroqConfig):
+        ModelRegistryHelper.__init__(self, model_aliases=_MODEL_ALIASES)
+        self._config = config
+
+    def completion(
+        self,
+        model_id: str,
+        content: InterleavedContent,
+        sampling_params: Optional[SamplingParams] = SamplingParams(),
+        response_format: Optional[ResponseFormat] = None,
+        stream: Optional[bool] = False,
+        logprobs: Optional[LogProbConfig] = None,
+    ) -> Union[CompletionResponse, AsyncIterator[CompletionResponseStreamChunk]]:
+        # Groq doesn't support non-chat completion as of time of writing
+        raise NotImplementedError()
+
+    async def chat_completion(
+        self,
+        model_id: str,
+        messages: List[Message],
+        sampling_params: Optional[SamplingParams] = SamplingParams(),
+        response_format: Optional[ResponseFormat] = None,
+        tools: Optional[List[ToolDefinition]] = None,
+        tool_choice: Optional[ToolChoice] = ToolChoice.auto,
+        tool_prompt_format: Optional[
+            ToolPromptFormat
+        ] = None,  # API default is ToolPromptFormat.json, we default to None to detect user input
+        stream: Optional[bool] = False,
+        logprobs: Optional[LogProbConfig] = None,
+    ) -> Union[
+        ChatCompletionResponse, AsyncIterator[ChatCompletionResponseStreamChunk]
+    ]:
+        model_id = self.get_provider_model_id(model_id)
+        if model_id == "llama-3.2-3b-preview":
+            warnings.warn(
+                "Groq only contains a preview version for llama-3.2-3b-instruct. "
+                "Preview models aren't recommended for production use. "
+                "They can be discontinued on short notice."
+            )
+
+        request = convert_chat_completion_request(
+            request=ChatCompletionRequest(
+                model=model_id,
+                messages=messages,
+                sampling_params=sampling_params,
+                response_format=response_format,
+                tools=tools,
+                tool_choice=tool_choice,
+                tool_prompt_format=tool_prompt_format,
+                stream=stream,
+                logprobs=logprobs,
+            )
+        )
+
+        try:
+            response = self._get_client().chat.completions.create(**request)
+        except groq.BadRequestError as e:
+            if e.body.get("error", {}).get("code") == "tool_use_failed":
+                # For smaller models, Groq may fail to call a tool even when the request is well formed
+                raise ValueError(
+                    "Groq failed to call a tool", e.body.get("error", {})
+                ) from e
+            else:
+                raise e
+
+        if stream:
+            return convert_chat_completion_response_stream(response)
+        else:
+            return convert_chat_completion_response(response)
+
+    async def embeddings(
+        self,
+        model_id: str,
+        contents: List[InterleavedContent],
+    ) -> EmbeddingsResponse:
+        raise NotImplementedError()
+
+    def _get_client(self) -> Groq:
+        if self._config.api_key is not None:
+            return Groq(api_key=self.config.api_key)
+        else:
+            provider_data = self.get_request_provider_data()
+            if provider_data is None or not provider_data.groq_api_key:
+                raise ValueError(
+                    'Pass Groq API Key in the header X-LlamaStack-ProviderData as { "groq_api_key": "<your api key>" }'
+                )
+            return Groq(api_key=provider_data.groq_api_key)