feat: add api.llama provider, llama-guard-4 model (#2058)

This PR adds a llama-stack inference provider for `api.llama.com`, as
well as adds entries for Llama-Guard-4 and updated Prompt-Guard models.

llama_stack/providers/registry/inference.py

@@ -227,6 +227,16 @@ def available_providers() -> List[ProviderSpec]:
                 provider_data_validator="llama_stack.providers.remote.inference.fireworks_openai_compat.config.FireworksProviderDataValidator",
             ),
         ),
+        remote_provider_spec(
+            api=Api.inference,
+            adapter=AdapterSpec(
+                adapter_type="llama-openai-compat",
+                pip_packages=["litellm"],
+                module="llama_stack.providers.remote.inference.llama_openai_compat",
+                config_class="llama_stack.providers.remote.inference.llama_openai_compat.config.LlamaCompatConfig",
+                provider_data_validator="llama_stack.providers.remote.inference.llama_openai_compat.config.LlamaProviderDataValidator",
+            ),
+        ),
         remote_provider_spec(
             api=Api.inference,
             adapter=AdapterSpec(

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

@@ -0,0 +1,17 @@
+# 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 llama_stack.apis.inference import Inference
+
+from .config import LlamaCompatConfig
+
+
+async def get_adapter_impl(config: LlamaCompatConfig, _deps) -> Inference:
+    # import dynamically so the import is used only when it is needed
+    from .llama import LlamaCompatInferenceAdapter
+
+    adapter = LlamaCompatInferenceAdapter(config)
+    return adapter

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

@@ -0,0 +1,38 @@
+# 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 Any, Dict, Optional
+
+from pydantic import BaseModel, Field
+
+from llama_stack.schema_utils import json_schema_type
+
+
+class LlamaProviderDataValidator(BaseModel):
+    llama_api_key: Optional[str] = Field(
+        default=None,
+        description="API key for api.llama models",
+    )
+
+
+@json_schema_type
+class LlamaCompatConfig(BaseModel):
+    api_key: Optional[str] = Field(
+        default=None,
+        description="The Llama API key",
+    )
+
+    openai_compat_api_base: str = Field(
+        default="https://api.llama.com/compat/v1/",
+        description="The URL for the Llama API server",
+    )
+
+    @classmethod
+    def sample_run_config(cls, api_key: str = "${env.LLAMA_API_KEY}", **kwargs) -> Dict[str, Any]:
+        return {
+            "openai_compat_api_base": "https://api.llama.com/compat/v1/",
+            "api_key": api_key,
+        }

llama_stack/providers/remote/inference/llama_openai_compat/llama.py

@@ -0,0 +1,34 @@
+# 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 llama_stack.providers.remote.inference.llama_openai_compat.config import (
+    LlamaCompatConfig,
+)
+from llama_stack.providers.utils.inference.litellm_openai_mixin import (
+    LiteLLMOpenAIMixin,
+)
+
+from .models import MODEL_ENTRIES
+
+
+class LlamaCompatInferenceAdapter(LiteLLMOpenAIMixin):
+    _config: LlamaCompatConfig
+
+    def __init__(self, config: LlamaCompatConfig):
+        LiteLLMOpenAIMixin.__init__(
+            self,
+            model_entries=MODEL_ENTRIES,
+            api_key_from_config=config.api_key,
+            provider_data_api_key_field="llama_api_key",
+            openai_compat_api_base=config.openai_compat_api_base,
+        )
+        self.config = config
+
+    async def initialize(self):
+        await super().initialize()
+
+    async def shutdown(self):
+        await super().shutdown()

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

@@ -0,0 +1,25 @@
+# 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 llama_stack.models.llama.sku_types import CoreModelId
+from llama_stack.providers.utils.inference.model_registry import (
+    build_hf_repo_model_entry,
+)
+
+MODEL_ENTRIES = [
+    build_hf_repo_model_entry(
+        "Llama-3.3-70B-Instruct",
+        CoreModelId.llama3_3_70b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "Llama-4-Scout-17B-16E-Instruct-FP8",
+        CoreModelId.llama4_scout_17b_16e_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "Llama-4-Maverick-17B-128E-Instruct-FP8",
+        CoreModelId.llama4_maverick_17b_128e_instruct.value,
+    ),
+]

llama_stack/providers/utils/inference/litellm_openai_mixin.py

@@ -90,6 +90,9 @@ class LiteLLMOpenAIMixin(
             raise ValueError(f"Unsupported model: {model.provider_resource_id}")
         return model
 
+    def get_litellm_model_name(self, model_id: str) -> str:
+        return "openai/" + model_id if self.is_openai_compat else model_id
+
     async def completion(
         self,
         model_id: str,
@@ -130,8 +133,7 @@ class LiteLLMOpenAIMixin(
         )
 
         params = await self._get_params(request)
-        if self.is_openai_compat:
-            params["model"] = "openai/" + params["model"]
+        params["model"] = self.get_litellm_model_name(params["model"])
 
         logger.debug(f"params to litellm (openai compat): {params}")
         # unfortunately, we need to use synchronous litellm.completion here because litellm
@@ -220,21 +222,23 @@ class LiteLLMOpenAIMixin(
                     else request.tool_config.tool_choice
                 )
 
+        return {
+            "model": request.model,
+            "api_key": self.get_api_key(),
+            "api_base": self.api_base,
+            **input_dict,
+            "stream": request.stream,
+            **get_sampling_options(request.sampling_params),
+        }
+
+    def get_api_key(self) -> str:
         provider_data = self.get_request_provider_data()
         key_field = self.provider_data_api_key_field
         if provider_data and getattr(provider_data, key_field, None):
             api_key = getattr(provider_data, key_field)
         else:
             api_key = self.api_key_from_config
-
-        return {
-            "model": request.model,
-            "api_key": api_key,
-            "api_base": self.api_base,
-            **input_dict,
-            "stream": request.stream,
-            **get_sampling_options(request.sampling_params),
-        }
+        return api_key
 
     async def embeddings(
         self,
@@ -247,7 +251,7 @@ class LiteLLMOpenAIMixin(
         model = await self.model_store.get_model(model_id)
 
         response = litellm.embedding(
-            model=model.provider_resource_id,
+            model=self.get_litellm_model_name(model.provider_resource_id),
             input=[interleaved_content_as_str(content) for content in contents],
         )
 
@@ -278,7 +282,7 @@ class LiteLLMOpenAIMixin(
     ) -> OpenAICompletion:
         model_obj = await self.model_store.get_model(model)
         params = await prepare_openai_completion_params(
-            model=model_obj.provider_resource_id,
+            model=self.get_litellm_model_name(model_obj.provider_resource_id),
             prompt=prompt,
             best_of=best_of,
             echo=echo,
@@ -297,6 +301,8 @@ class LiteLLMOpenAIMixin(
             user=user,
             guided_choice=guided_choice,
             prompt_logprobs=prompt_logprobs,
+            api_key=self.get_api_key(),
+            api_base=self.api_base,
         )
         return await litellm.atext_completion(**params)
 
@@ -328,7 +334,7 @@ class LiteLLMOpenAIMixin(
     ) -> Union[OpenAIChatCompletion, AsyncIterator[OpenAIChatCompletionChunk]]:
         model_obj = await self.model_store.get_model(model)
         params = await prepare_openai_completion_params(
-            model=model_obj.provider_resource_id,
+            model=self.get_litellm_model_name(model_obj.provider_resource_id),
             messages=messages,
             frequency_penalty=frequency_penalty,
             function_call=function_call,
@@ -351,6 +357,8 @@ class LiteLLMOpenAIMixin(
             top_logprobs=top_logprobs,
             top_p=top_p,
             user=user,
+            api_key=self.get_api_key(),
+            api_base=self.api_base,
         )
         return await litellm.acompletion(**params)
 

llama_stack/providers/utils/inference/openai_compat.py

@@ -638,10 +638,13 @@ async def convert_message_to_openai_dict_new(
             )
             for tool in message.tool_calls
         ]
+        params = {}
+        if tool_calls:
+            params["tool_calls"] = tool_calls
         out = OpenAIChatCompletionAssistantMessage(
             role="assistant",
             content=await _convert_message_content(message.content),
-            tool_calls=tool_calls or None,
+            **params,
         )
     elif isinstance(message, ToolResponseMessage):
         out = OpenAIChatCompletionToolMessage(