phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-27 06:28:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'main' into allow-dynamic-models-ollama

Browse source
This commit is contained in:
Matthew Farrellee 2025-07-14 17:29:18 -04:00
commit b6a334604c
113 changed files with 3795 additions and 3100 deletions
Download patch file Download diff file Expand all files Collapse all files

39
llama_stack/providers/remote/inference/anthropic/models.py
View file

@ -15,21 +15,26 @@ LLM_MODEL_IDS = [
"anthropic/claude-3-5-haiku-latest",
]
SAFETY_MODELS_ENTRIES = []
MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
ProviderModelEntry(
provider_model_id="anthropic/voyage-3",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 1024, "context_length": 32000},
),
ProviderModelEntry(
provider_model_id="anthropic/voyage-3-lite",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 512, "context_length": 32000},
),
ProviderModelEntry(
provider_model_id="anthropic/voyage-code-3",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 1024, "context_length": 32000},
),
]
MODEL_ENTRIES = (
[ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+ [
ProviderModelEntry(
provider_model_id="anthropic/voyage-3",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 1024, "context_length": 32000},
),
ProviderModelEntry(
provider_model_id="anthropic/voyage-3-lite",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 512, "context_length": 32000},
),
ProviderModelEntry(
provider_model_id="anthropic/voyage-code-3",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 1024, "context_length": 32000},
),
]
+ SAFETY_MODELS_ENTRIES
)

6
llama_stack/providers/remote/inference/bedrock/models.py
View file

@ -9,6 +9,10 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = []
# https://docs.aws.amazon.com/bedrock/latest/userguide/models-supported.html
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"meta.llama3-1-8b-instruct-v1:0",
@ -22,4 +26,4 @@ MODEL_ENTRIES = [
"meta.llama3-1-405b-instruct-v1:0",
CoreModelId.llama3_1_405b_instruct.value,
),
]
] + SAFETY_MODELS_ENTRIES

9
llama_stack/providers/remote/inference/cerebras/models.py
View file

@ -9,6 +9,9 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = []
# https://inference-docs.cerebras.ai/models
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"llama3.1-8b",
@ -18,4 +21,8 @@ MODEL_ENTRIES = [
"llama-3.3-70b",
CoreModelId.llama3_3_70b_instruct.value,
),
]
build_hf_repo_model_entry(
"llama-4-scout-17b-16e-instruct",
CoreModelId.llama4_scout_17b_16e_instruct.value,
),
] + SAFETY_MODELS_ENTRIES

9
llama_stack/providers/remote/inference/databricks/databricks.py
View file

@ -47,7 +47,10 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
from .config import DatabricksImplConfig
model_entries = [
SAFETY_MODELS_ENTRIES = []
# https://docs.databricks.com/aws/en/machine-learning/model-serving/foundation-model-overview
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"databricks-meta-llama-3-1-70b-instruct",
CoreModelId.llama3_1_70b_instruct.value,
@ -56,7 +59,7 @@ model_entries = [
"databricks-meta-llama-3-1-405b-instruct",
CoreModelId.llama3_1_405b_instruct.value,
),
]
] + SAFETY_MODELS_ENTRIES
class DatabricksInferenceAdapter(
@ -66,7 +69,7 @@ class DatabricksInferenceAdapter(
OpenAICompletionToLlamaStackMixin,
):
def __init__(self, config: DatabricksImplConfig) -> None:
ModelRegistryHelper.__init__(self, model_entries=model_entries)
ModelRegistryHelper.__init__(self, model_entries=MODEL_ENTRIES)
self.config = config
async def initialize(self) -> None:

21
llama_stack/providers/remote/inference/fireworks/models.py
View file

@ -11,6 +11,17 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = [
build_hf_repo_model_entry(
"accounts/fireworks/models/llama-guard-3-8b",
CoreModelId.llama_guard_3_8b.value,
),
build_hf_repo_model_entry(
"accounts/fireworks/models/llama-guard-3-11b-vision",
CoreModelId.llama_guard_3_11b_vision.value,
),
]
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"accounts/fireworks/models/llama-v3p1-8b-instruct",
@ -40,14 +51,6 @@ MODEL_ENTRIES = [
"accounts/fireworks/models/llama-v3p3-70b-instruct",
CoreModelId.llama3_3_70b_instruct.value,
),
build_hf_repo_model_entry(
"accounts/fireworks/models/llama-guard-3-8b",
CoreModelId.llama_guard_3_8b.value,
),
build_hf_repo_model_entry(
"accounts/fireworks/models/llama-guard-3-11b-vision",
CoreModelId.llama_guard_3_11b_vision.value,
),
build_hf_repo_model_entry(
"accounts/fireworks/models/llama4-scout-instruct-basic",
CoreModelId.llama4_scout_17b_16e_instruct.value,
@ -64,4 +67,4 @@ MODEL_ENTRIES = [
"context_length": 8192,
},
),
]
] + SAFETY_MODELS_ENTRIES

19
llama_stack/providers/remote/inference/gemini/models.py
View file

@ -17,11 +17,16 @@ LLM_MODEL_IDS = [
"gemini/gemini-2.5-pro",
]
SAFETY_MODELS_ENTRIES = []
MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
ProviderModelEntry(
provider_model_id="gemini/text-embedding-004",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 768, "context_length": 2048},
),
]
MODEL_ENTRIES = (
[ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+ [
ProviderModelEntry(
provider_model_id="gemini/text-embedding-004",
model_type=ModelType.embedding,
metadata={"embedding_dimension": 768, "context_length": 2048},
),
]
+ SAFETY_MODELS_ENTRIES
)

14
llama_stack/providers/remote/inference/groq/groq.py
View file

@ -38,24 +38,18 @@ class GroqInferenceAdapter(LiteLLMOpenAIMixin):
provider_data_api_key_field="groq_api_key",
)
self.config = config
self._openai_client = None
async def initialize(self):
await super().initialize()
async def shutdown(self):
await super().shutdown()
if self._openai_client:
await self._openai_client.close()
self._openai_client = None
def _get_openai_client(self) -> AsyncOpenAI:
if not self._openai_client:
self._openai_client = AsyncOpenAI(
base_url=f"{self.config.url}/openai/v1",
api_key=self.config.api_key,
)
return self._openai_client
return AsyncOpenAI(
base_url=f"{self.config.url}/openai/v1",
api_key=self.get_api_key(),
)
async def openai_chat_completion(
self,

4
llama_stack/providers/remote/inference/groq/models.py
View file

@ -10,6 +10,8 @@ from llama_stack.providers.utils.inference.model_registry import (
build_model_entry,
)
SAFETY_MODELS_ENTRIES = []
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"groq/llama3-8b-8192",
@ -51,4 +53,4 @@ MODEL_ENTRIES = [
"groq/meta-llama/llama-4-maverick-17b-128e-instruct",
CoreModelId.llama4_maverick_17b_128e_instruct.value,
),
]
] + SAFETY_MODELS_ENTRIES

5
llama_stack/providers/remote/inference/nvidia/models.py
View file

@ -11,6 +11,9 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = []
# https://docs.nvidia.com/nim/large-language-models/latest/supported-llm-agnostic-architectures.html
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"meta/llama3-8b-instruct",
@ -99,4 +102,4 @@ MODEL_ENTRIES = [
),
# TODO(mf): how do we handle Nemotron models?
# "Llama3.1-Nemotron-51B-Instruct" -> "meta/llama-3.1-nemotron-51b-instruct",
]
] + SAFETY_MODELS_ENTRIES

28
llama_stack/providers/remote/inference/openai/models.py
View file

@ -48,16 +48,20 @@ EMBEDDING_MODEL_IDS: dict[str, EmbeddingModelInfo] = {
"text-embedding-3-small": EmbeddingModelInfo(1536, 8192),
"text-embedding-3-large": EmbeddingModelInfo(3072, 8192),
}
SAFETY_MODELS_ENTRIES = []
MODEL_ENTRIES = [ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS] + [
ProviderModelEntry(
provider_model_id=model_id,
model_type=ModelType.embedding,
metadata={
"embedding_dimension": model_info.embedding_dimension,
"context_length": model_info.context_length,
},
)
for model_id, model_info in EMBEDDING_MODEL_IDS.items()
]
MODEL_ENTRIES = (
[ProviderModelEntry(provider_model_id=m) for m in LLM_MODEL_IDS]
+ [
ProviderModelEntry(
provider_model_id=model_id,
model_type=ModelType.embedding,
metadata={
"embedding_dimension": model_info.embedding_dimension,
"context_length": model_info.context_length,
},
)
for model_id, model_info in EMBEDDING_MODEL_IDS.items()
]
+ SAFETY_MODELS_ENTRIES
)

14
llama_stack/providers/remote/inference/openai/openai.py
View file

@ -59,9 +59,6 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
# if we do not set this, users will be exposed to the
# litellm specific model names, an abstraction leak.
self.is_openai_compat = True
self._openai_client = AsyncOpenAI(
api_key=self.config.api_key,
)
async def initialize(self) -> None:
await super().initialize()
@ -69,6 +66,11 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
async def shutdown(self) -> None:
await super().shutdown()
def _get_openai_client(self) -> AsyncOpenAI:
return AsyncOpenAI(
api_key=self.get_api_key(),
)
async def openai_completion(
self,
model: str,
@ -120,7 +122,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
user=user,
suffix=suffix,
)
return await self._openai_client.completions.create(**params)
return await self._get_openai_client().completions.create(**params)
async def openai_chat_completion(
self,
@ -176,7 +178,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
top_p=top_p,
user=user,
)
return await self._openai_client.chat.completions.create(**params)
return await self._get_openai_client().chat.completions.create(**params)
async def openai_embeddings(
self,
@ -204,7 +206,7 @@ class OpenAIInferenceAdapter(LiteLLMOpenAIMixin):
params["user"] = user
# Call OpenAI embeddings API
response = await self._openai_client.embeddings.create(**params)
response = await self._get_openai_client().embeddings.create(**params)
data = []
for i, embedding_data in enumerate(response.data):

12
llama_stack/providers/remote/inference/runpod/runpod.py
View file

@ -11,7 +11,7 @@ from llama_stack.apis.inference import * # noqa: F403
from llama_stack.apis.inference import OpenAIEmbeddingsResponse
# from llama_stack.providers.datatypes import ModelsProtocolPrivate
from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper, build_hf_repo_model_entry
from llama_stack.providers.utils.inference.openai_compat import (
OpenAIChatCompletionToLlamaStackMixin,
OpenAICompletionToLlamaStackMixin,
@ -25,6 +25,8 @@ from llama_stack.providers.utils.inference.prompt_adapter import (
from .config import RunpodImplConfig
# https://docs.runpod.io/serverless/vllm/overview#compatible-models
# https://github.com/runpod-workers/worker-vllm/blob/main/README.md#compatible-model-architectures
RUNPOD_SUPPORTED_MODELS = {
"Llama3.1-8B": "meta-llama/Llama-3.1-8B",
"Llama3.1-70B": "meta-llama/Llama-3.1-70B",
@ -40,6 +42,14 @@ RUNPOD_SUPPORTED_MODELS = {
"Llama3.2-3B": "meta-llama/Llama-3.2-3B",
}
SAFETY_MODELS_ENTRIES = []
# Create MODEL_ENTRIES from RUNPOD_SUPPORTED_MODELS for compatibility with starter template
MODEL_ENTRIES = [
build_hf_repo_model_entry(provider_model_id, model_descriptor)
for provider_model_id, model_descriptor in RUNPOD_SUPPORTED_MODELS.items()
] + SAFETY_MODELS_ENTRIES
class RunpodInferenceAdapter(
ModelRegistryHelper,

14
llama_stack/providers/remote/inference/sambanova/models.py
View file

@ -9,6 +9,14 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = [
build_hf_repo_model_entry(
"sambanova/Meta-Llama-Guard-3-8B",
CoreModelId.llama_guard_3_8b.value,
),
]
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"sambanova/Meta-Llama-3.1-8B-Instruct",
@ -46,8 +54,4 @@ MODEL_ENTRIES = [
"sambanova/Llama-4-Maverick-17B-128E-Instruct",
CoreModelId.llama4_maverick_17b_128e_instruct.value,
),
build_hf_repo_model_entry(
"sambanova/Meta-Llama-Guard-3-8B",
CoreModelId.llama_guard_3_8b.value,
),
]
] + SAFETY_MODELS_ENTRIES

21
llama_stack/providers/remote/inference/sambanova/sambanova.py
View file

@ -7,6 +7,7 @@
import json
from collections.abc import Iterable
import requests
from openai.types.chat import (
ChatCompletionAssistantMessageParam as OpenAIChatCompletionAssistantMessage,
)
@ -56,6 +57,7 @@ from llama_stack.apis.inference import (
ToolResponseMessage,
UserMessage,
)
from llama_stack.apis.models import Model
from llama_stack.log import get_logger
from llama_stack.models.llama.datatypes import BuiltinTool
from llama_stack.providers.utils.inference.litellm_openai_mixin import LiteLLMOpenAIMixin
@ -176,10 +178,11 @@ class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
def __init__(self, config: SambaNovaImplConfig):
self.config = config
self.environment_available_models = []
LiteLLMOpenAIMixin.__init__(
self,
model_entries=MODEL_ENTRIES,
api_key_from_config=self.config.api_key,
api_key_from_config=self.config.api_key.get_secret_value() if self.config.api_key else None,
provider_data_api_key_field="sambanova_api_key",
)
@ -246,6 +249,22 @@ class SambaNovaInferenceAdapter(LiteLLMOpenAIMixin):
**get_sampling_options(request.sampling_params),
}
async def register_model(self, model: Model) -> Model:
model_id = self.get_provider_model_id(model.provider_resource_id)
list_models_url = self.config.url + "/models"
if len(self.environment_available_models) == 0:
try:
response = requests.get(list_models_url)
response.raise_for_status()
except requests.exceptions.RequestException as e:
raise RuntimeError(f"Request to {list_models_url} failed") from e
self.environment_available_models = [model.get("id") for model in response.json().get("data", {})]
if model_id.split("sambanova/")[-1] not in self.environment_available_models:
logger.warning(f"Model {model_id} not available in {list_models_url}")
return model
async def initialize(self):
await super().initialize()

20
llama_stack/providers/remote/inference/together/models.py
View file

@ -11,6 +11,16 @@ from llama_stack.providers.utils.inference.model_registry import (
build_hf_repo_model_entry,
)
SAFETY_MODELS_ENTRIES = [
build_hf_repo_model_entry(
"meta-llama/Llama-Guard-3-8B",
CoreModelId.llama_guard_3_8b.value,
),
build_hf_repo_model_entry(
"meta-llama/Llama-Guard-3-11B-Vision-Turbo",
CoreModelId.llama_guard_3_11b_vision.value,
),
]
MODEL_ENTRIES = [
build_hf_repo_model_entry(
"meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo",
@ -40,14 +50,6 @@ MODEL_ENTRIES = [
"meta-llama/Llama-3.3-70B-Instruct-Turbo",
CoreModelId.llama3_3_70b_instruct.value,
),
build_hf_repo_model_entry(
"meta-llama/Meta-Llama-Guard-3-8B",
CoreModelId.llama_guard_3_8b.value,
),
build_hf_repo_model_entry(
"meta-llama/Llama-Guard-3-11B-Vision-Turbo",
CoreModelId.llama_guard_3_11b_vision.value,
),
ProviderModelEntry(
provider_model_id="togethercomputer/m2-bert-80M-8k-retrieval",
model_type=ModelType.embedding,
@ -78,4 +80,4 @@ MODEL_ENTRIES = [
"together/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8",
],
),
]
] + SAFETY_MODELS_ENTRIES

47
llama_stack/providers/remote/inference/together/together.py
View file

@ -68,19 +68,12 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
def __init__(self, config: TogetherImplConfig) -> None:
ModelRegistryHelper.__init__(self, MODEL_ENTRIES)
self.config = config
self._client = None
self._openai_client = None
async def initialize(self) -> None:
pass
async def shutdown(self) -> None:
if self._client:
# Together client has no close method, so just set to None
self._client = None
if self._openai_client:
await self._openai_client.close()
self._openai_client = None
pass
async def completion(
self,
@ -108,29 +101,25 @@ class TogetherInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProvi
return await self._nonstream_completion(request)
def _get_client(self) -> AsyncTogether:
if not self._client:
together_api_key = None
config_api_key = self.config.api_key.get_secret_value() if self.config.api_key else None
if config_api_key:
together_api_key = config_api_key
else:
provider_data = self.get_request_provider_data()
if provider_data is None or not provider_data.together_api_key:
raise ValueError(
'Pass Together API Key in the header X-LlamaStack-Provider-Data as { "together_api_key": <your api key>}'
)
together_api_key = provider_data.together_api_key
self._client = AsyncTogether(api_key=together_api_key)
return self._client
together_api_key = None
config_api_key = self.config.api_key.get_secret_value() if self.config.api_key else None
if config_api_key:
together_api_key = config_api_key
else:
provider_data = self.get_request_provider_data()
if provider_data is None or not provider_data.together_api_key:
raise ValueError(
'Pass Together API Key in the header X-LlamaStack-Provider-Data as { "together_api_key": <your api key>}'
)
together_api_key = provider_data.together_api_key
return AsyncTogether(api_key=together_api_key)
def _get_openai_client(self) -> AsyncOpenAI:
if not self._openai_client:
together_client = self._get_client().client
self._openai_client = AsyncOpenAI(
base_url=together_client.base_url,
api_key=together_client.api_key,
)
return self._openai_client
together_client = self._get_client().client
return AsyncOpenAI(
base_url=together_client.base_url,
api_key=together_client.api_key,
)
async def _nonstream_completion(self, request: CompletionRequest) -> ChatCompletionResponse:
params = await self._get_params(request)