phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2026-01-02 02:10:00 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

todos

Browse source
This commit is contained in:
Xi Yan 2025-03-18 21:49:11 -07:00
parent 011fd59a29
commit 8a576d7d72
24 changed files with 297 additions and 2525 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama_stack/distribution/routers/__init__.py
View file

@ -44,9 +44,7 @@ async def get_routing_table_impl(
return impl
async def get_auto_router_impl(
api: Api, routing_table: RoutingTable, deps: Dict[str, Any]
) -> Any:
async def get_auto_router_impl(api: Api, routing_table: RoutingTable, deps: Dict[str, Any]) -> Any:
from .routers import (
DatasetIORouter,
EvalRouter,

128
llama_stack/distribution/routers/routers.py
View file

@ -8,19 +8,12 @@ import time
from typing import Any, AsyncGenerator, AsyncIterator, Dict, List, Optional, Union
from llama_stack.apis.common.content_types import (
URL,
InterleavedContent,
InterleavedContentItem,
URL,
)
from llama_stack.apis.datasetio import DatasetIO, IterrowsResponse
from llama_stack.apis.datasets import DatasetPurpose, DataSource
from llama_stack.apis.eval import (
BenchmarkConfig,
Eval,
EvaluateResponse,
Job,
JobStatus,
)
from llama_stack.apis.inference import (
ChatCompletionResponse,
ChatCompletionResponseEventType,
@ -42,12 +35,6 @@ from llama_stack.apis.inference import (
)
from llama_stack.apis.models import Model, ModelType
from llama_stack.apis.safety import RunShieldResponse, Safety
from llama_stack.apis.scoring import (
ScoreBatchResponse,
ScoreResponse,
Scoring,
ScoringFnParams,
)
from llama_stack.apis.shields import Shield
from llama_stack.apis.telemetry import MetricEvent, MetricInResponse, Telemetry
from llama_stack.apis.tools import (
@ -94,9 +81,7 @@ class VectorIORouter(VectorIO):
provider_id: Optional[str] = None,
provider_vector_db_id: Optional[str] = None,
) -> None:
logger.debug(
f"VectorIORouter.register_vector_db: {vector_db_id}, {embedding_model}"
)
logger.debug(f"VectorIORouter.register_vector_db: {vector_db_id}, {embedding_model}")
await self.routing_table.register_vector_db(
vector_db_id,
embedding_model,
@ -114,9 +99,7 @@ class VectorIORouter(VectorIO):
logger.debug(
f"VectorIORouter.insert_chunks: {vector_db_id}, {len(chunks)} chunks, ttl_seconds={ttl_seconds}, chunk_ids={[chunk.metadata['document_id'] for chunk in chunks[:3]]}{' and more...' if len(chunks) > 3 else ''}",
)
return await self.routing_table.get_provider_impl(vector_db_id).insert_chunks(
vector_db_id, chunks, ttl_seconds
)
return await self.routing_table.get_provider_impl(vector_db_id).insert_chunks(vector_db_id, chunks, ttl_seconds)
async def query_chunks(
self,
@ -125,9 +108,7 @@ class VectorIORouter(VectorIO):
params: Optional[Dict[str, Any]] = None,
) -> QueryChunksResponse:
logger.debug(f"VectorIORouter.query_chunks: {vector_db_id}")
return await self.routing_table.get_provider_impl(vector_db_id).query_chunks(
vector_db_id, query, params
)
return await self.routing_table.get_provider_impl(vector_db_id).query_chunks(vector_db_id, query, params)
class InferenceRouter(Inference):
@ -164,9 +145,7 @@ class InferenceRouter(Inference):
logger.debug(
f"InferenceRouter.register_model: {model_id=} {provider_model_id=} {provider_id=} {metadata=} {model_type=}",
)
await self.routing_table.register_model(
model_id, provider_model_id, provider_id, metadata, model_type
)
await self.routing_table.register_model(model_id, provider_model_id, provider_id, metadata, model_type)
def _construct_metrics(
self,
@ -220,16 +199,11 @@ class InferenceRouter(Inference):
total_tokens: int,
model: Model,
) -> List[MetricInResponse]:
metrics = self._construct_metrics(
prompt_tokens, completion_tokens, total_tokens, model
)
metrics = self._construct_metrics(prompt_tokens, completion_tokens, total_tokens, model)
if self.telemetry:
for metric in metrics:
await self.telemetry.log_event(metric)
return [
MetricInResponse(metric=metric.metric, value=metric.value)
for metric in metrics
]
return [MetricInResponse(metric=metric.metric, value=metric.value) for metric in metrics]
async def _count_tokens(
self,
@ -254,9 +228,7 @@ class InferenceRouter(Inference):
stream: Optional[bool] = False,
logprobs: Optional[LogProbConfig] = None,
tool_config: Optional[ToolConfig] = None,
) -> Union[
ChatCompletionResponse, AsyncIterator[ChatCompletionResponseStreamChunk]
]:
) -> Union[ChatCompletionResponse, AsyncIterator[ChatCompletionResponseStreamChunk]]:
logger.debug(
f"InferenceRouter.chat_completion: {model_id=}, {stream=}, {messages=}, {tools=}, {tool_config=}, {response_format=}",
)
@ -266,19 +238,12 @@ class InferenceRouter(Inference):
if model is None:
raise ValueError(f"Model '{model_id}' not found")
if model.model_type == ModelType.embedding:
raise ValueError(
f"Model '{model_id}' is an embedding model and does not support chat completions"
)
raise ValueError(f"Model '{model_id}' is an embedding model and does not support chat completions")
if tool_config:
if tool_choice and tool_choice != tool_config.tool_choice:
raise ValueError("tool_choice and tool_config.tool_choice must match")
if (
tool_prompt_format
and tool_prompt_format != tool_config.tool_prompt_format
):
raise ValueError(
"tool_prompt_format and tool_config.tool_prompt_format must match"
)
if tool_prompt_format and tool_prompt_format != tool_config.tool_prompt_format:
raise ValueError("tool_prompt_format and tool_config.tool_prompt_format must match")
else:
params = {}
if tool_choice:
@ -296,14 +261,9 @@ class InferenceRouter(Inference):
pass
else:
# verify tool_choice is one of the tools
tool_names = [
t.tool_name if isinstance(t.tool_name, str) else t.tool_name.value
for t in tools
]
tool_names = [t.tool_name if isinstance(t.tool_name, str) else t.tool_name.value for t in tools]
if tool_config.tool_choice not in tool_names:
raise ValueError(
f"Tool choice {tool_config.tool_choice} is not one of the tools: {tool_names}"
)
raise ValueError(f"Tool choice {tool_config.tool_choice} is not one of the tools: {tool_names}")
params = dict(
model_id=model_id,
@ -318,25 +278,17 @@ class InferenceRouter(Inference):
tool_config=tool_config,
)
provider = self.routing_table.get_provider_impl(model_id)
prompt_tokens = await self._count_tokens(
messages, tool_config.tool_prompt_format
)
prompt_tokens = await self._count_tokens(messages, tool_config.tool_prompt_format)
if stream:
async def stream_generator():
completion_text = ""
async for chunk in await provider.chat_completion(**params):
if (
chunk.event.event_type
== ChatCompletionResponseEventType.progress
):
if chunk.event.event_type == ChatCompletionResponseEventType.progress:
if chunk.event.delta.type == "text":
completion_text += chunk.event.delta.text
if (
chunk.event.event_type
== ChatCompletionResponseEventType.complete
):
if chunk.event.event_type == ChatCompletionResponseEventType.complete:
completion_tokens = await self._count_tokens(
[
CompletionMessage(
@ -353,11 +305,7 @@ class InferenceRouter(Inference):
total_tokens,
model,
)
chunk.metrics = (
metrics
if chunk.metrics is None
else chunk.metrics + metrics
)
chunk.metrics = metrics if chunk.metrics is None else chunk.metrics + metrics
yield chunk
return stream_generator()
@ -374,9 +322,7 @@ class InferenceRouter(Inference):
total_tokens,
model,
)
response.metrics = (
metrics if response.metrics is None else response.metrics + metrics
)
response.metrics = metrics if response.metrics is None else response.metrics + metrics
return response
async def completion(
@ -397,9 +343,7 @@ class InferenceRouter(Inference):
if model is None:
raise ValueError(f"Model '{model_id}' not found")
if model.model_type == ModelType.embedding:
raise ValueError(
f"Model '{model_id}' is an embedding model and does not support chat completions"
)
raise ValueError(f"Model '{model_id}' is an embedding model and does not support chat completions")
provider = self.routing_table.get_provider_impl(model_id)
params = dict(
model_id=model_id,
@ -419,11 +363,7 @@ class InferenceRouter(Inference):
async for chunk in await provider.completion(**params):
if hasattr(chunk, "delta"):
completion_text += chunk.delta
if (
hasattr(chunk, "stop_reason")
and chunk.stop_reason
and self.telemetry
):
if hasattr(chunk, "stop_reason") and chunk.stop_reason and self.telemetry:
completion_tokens = await self._count_tokens(completion_text)
total_tokens = (prompt_tokens or 0) + (completion_tokens or 0)
metrics = await self._compute_and_log_token_usage(
@ -432,11 +372,7 @@ class InferenceRouter(Inference):
total_tokens,
model,
)
chunk.metrics = (
metrics
if chunk.metrics is None
else chunk.metrics + metrics
)
chunk.metrics = metrics if chunk.metrics is None else chunk.metrics + metrics
yield chunk
return stream_generator()
@ -450,9 +386,7 @@ class InferenceRouter(Inference):
total_tokens,
model,
)
response.metrics = (
metrics if response.metrics is None else response.metrics + metrics
)
response.metrics = metrics if response.metrics is None else response.metrics + metrics
return response
async def embeddings(
@ -468,9 +402,7 @@ class InferenceRouter(Inference):
if model is None:
raise ValueError(f"Model '{model_id}' not found")
if model.model_type == ModelType.llm:
raise ValueError(
f"Model '{model_id}' is an LLM model and does not support embeddings"
)
raise ValueError(f"Model '{model_id}' is an LLM model and does not support embeddings")
return await self.routing_table.get_provider_impl(model_id).embeddings(
model_id=model_id,
contents=contents,
@ -504,9 +436,7 @@ class SafetyRouter(Safety):
params: Optional[Dict[str, Any]] = None,
) -> Shield:
logger.debug(f"SafetyRouter.register_shield: {shield_id}")
return await self.routing_table.register_shield(
shield_id, provider_shield_id, provider_id, params
)
return await self.routing_table.register_shield(shield_id, provider_shield_id, provider_id, params)
async def run_shield(
self,
@ -607,9 +537,9 @@ class ToolRuntimeRouter(ToolRuntime):
logger.debug(
f"ToolRuntimeRouter.RagToolImpl.insert: {vector_db_id}, {len(documents)} documents, chunk_size={chunk_size_in_tokens}"
)
return await self.routing_table.get_provider_impl(
"insert_into_memory"
).insert(documents, vector_db_id, chunk_size_in_tokens)
return await self.routing_table.get_provider_impl("insert_into_memory").insert(
documents, vector_db_id, chunk_size_in_tokens
)
def __init__(
self,
@ -642,6 +572,4 @@ class ToolRuntimeRouter(ToolRuntime):
self, tool_group_id: Optional[str] = None, mcp_endpoint: Optional[URL] = None
) -> List[ToolDef]:
logger.debug(f"ToolRuntimeRouter.list_runtime_tools: {tool_group_id}")
return await self.routing_table.get_provider_impl(tool_group_id).list_tools(
tool_group_id, mcp_endpoint
)
return await self.routing_table.get_provider_impl(tool_group_id).list_tools(tool_group_id, mcp_endpoint)

53
llama_stack/distribution/routers/routing_tables.py
View file

@ -12,7 +12,6 @@ from pydantic import TypeAdapter
from llama_stack.apis.benchmarks import Benchmark, Benchmarks, ListBenchmarksResponse
from llama_stack.apis.common.content_types import URL
from llama_stack.apis.common.type_system import ParamType
from llama_stack.apis.datasets import (
Dataset,
DatasetPurpose,
@ -95,9 +94,7 @@ class CommonRoutingTableImpl(RoutingTable):
self.dist_registry = dist_registry
async def initialize(self) -> None:
async def add_objects(
objs: List[RoutableObjectWithProvider], provider_id: str, cls
) -> None:
async def add_objects(objs: List[RoutableObjectWithProvider], provider_id: str, cls) -> None:
for obj in objs:
if cls is None:
obj.provider_id = provider_id
@ -126,9 +123,7 @@ class CommonRoutingTableImpl(RoutingTable):
for p in self.impls_by_provider_id.values():
await p.shutdown()
def get_provider_impl(
self, routing_key: str, provider_id: Optional[str] = None
) -> Any:
def get_provider_impl(self, routing_key: str, provider_id: Optional[str] = None) -> Any:
def apiname_object():
if isinstance(self, ModelsRoutingTable):
return ("Inference", "model")
@ -164,9 +159,7 @@ class CommonRoutingTableImpl(RoutingTable):
raise ValueError(f"Provider not found for `{routing_key}`")
async def get_object_by_identifier(
self, type: str, identifier: str
) -> Optional[RoutableObjectWithProvider]:
async def get_object_by_identifier(self, type: str, identifier: str) -> Optional[RoutableObjectWithProvider]:
# Get from disk registry
obj = await self.dist_registry.get(type, identifier)
if not obj:
@ -176,13 +169,9 @@ class CommonRoutingTableImpl(RoutingTable):
async def unregister_object(self, obj: RoutableObjectWithProvider) -> None:
await self.dist_registry.delete(obj.type, obj.identifier)
await unregister_object_from_provider(
obj, self.impls_by_provider_id[obj.provider_id]
)
await unregister_object_from_provider(obj, self.impls_by_provider_id[obj.provider_id])
async def register_object(
self, obj: RoutableObjectWithProvider
) -> RoutableObjectWithProvider:
async def register_object(self, obj: RoutableObjectWithProvider) -> RoutableObjectWithProvider:
# if provider_id is not specified, pick an arbitrary one from existing entries
if not obj.provider_id and len(self.impls_by_provider_id) > 0:
obj.provider_id = list(self.impls_by_provider_id.keys())[0]
@ -240,9 +229,7 @@ class ModelsRoutingTable(CommonRoutingTableImpl, Models):
if model_type is None:
model_type = ModelType.llm
if "embedding_dimension" not in metadata and model_type == ModelType.embedding:
raise ValueError(
"Embedding model must have an embedding dimension in its metadata"
)
raise ValueError("Embedding model must have an embedding dimension in its metadata")
model = Model(
identifier=model_id,
provider_resource_id=provider_model_id,
@ -262,9 +249,7 @@ class ModelsRoutingTable(CommonRoutingTableImpl, Models):
class ShieldsRoutingTable(CommonRoutingTableImpl, Shields):
async def list_shields(self) -> ListShieldsResponse:
return ListShieldsResponse(
data=await self.get_all_with_type(ResourceType.shield.value)
)
return ListShieldsResponse(data=await self.get_all_with_type(ResourceType.shield.value))
async def get_shield(self, identifier: str) -> Shield:
shield = await self.get_object_by_identifier("shield", identifier)
@ -329,18 +314,14 @@ class VectorDBsRoutingTable(CommonRoutingTableImpl, VectorDBs):
f"No provider specified and multiple providers available. Arbitrarily selected the first provider {provider_id}."
)
else:
raise ValueError(
"No provider available. Please configure a vector_io provider."
)
raise ValueError("No provider available. Please configure a vector_io provider.")
model = await self.get_object_by_identifier("model", embedding_model)
if model is None:
raise ValueError(f"Model {embedding_model} not found")
if model.model_type != ModelType.embedding:
raise ValueError(f"Model {embedding_model} is not an embedding model")
if "embedding_dimension" not in model.metadata:
raise ValueError(
f"Model {embedding_model} does not have an embedding dimension"
)
raise ValueError(f"Model {embedding_model} does not have an embedding dimension")
vector_db_data = {
"identifier": vector_db_id,
"type": ResourceType.vector_db.value,
@ -362,9 +343,7 @@ class VectorDBsRoutingTable(CommonRoutingTableImpl, VectorDBs):
class DatasetsRoutingTable(CommonRoutingTableImpl, Datasets):
async def list_datasets(self) -> ListDatasetsResponse:
return ListDatasetsResponse(
data=await self.get_all_with_type(ResourceType.dataset.value)
)
return ListDatasetsResponse(data=await self.get_all_with_type(ResourceType.dataset.value))
async def get_dataset(self, dataset_id: str) -> Dataset:
dataset = await self.get_object_by_identifier("dataset", dataset_id)
@ -447,9 +426,7 @@ class BenchmarksRoutingTable(CommonRoutingTableImpl, Benchmarks):
# TODO (xiyan): we will need a way to infer provider_id for evaluation
# keep it as meta-reference for now
if len(self.impls_by_provider_id) == 0:
raise ValueError(
"No evaluation providers available. Please configure an evaluation provider."
)
raise ValueError("No evaluation providers available. Please configure an evaluation provider.")
provider_id = list(self.impls_by_provider_id.keys())[0]
benchmark = Benchmark(
@ -491,12 +468,8 @@ class ToolGroupsRoutingTable(CommonRoutingTableImpl, ToolGroups):
args: Optional[Dict[str, Any]] = None,
) -> None:
tools = []
tool_defs = await self.impls_by_provider_id[provider_id].list_runtime_tools(
toolgroup_id, mcp_endpoint
)
tool_host = (
ToolHost.model_context_protocol if mcp_endpoint else ToolHost.distribution
)
tool_defs = await self.impls_by_provider_id[provider_id].list_runtime_tools(toolgroup_id, mcp_endpoint)
tool_host = ToolHost.model_context_protocol if mcp_endpoint else ToolHost.distribution
for tool_def in tool_defs:
tools.append(