forked from phoenix-oss/llama-stack-mirror
[API Updates] Model / shield / memory-bank routing + agent persistence + support for private headers (#92)
This is yet another of those large PRs (hopefully we will have less and less of them as things mature fast). This one introduces substantial improvements and some simplifications to the stack. Most important bits: * Agents reference implementation now has support for session / turn persistence. The default implementation uses sqlite but there's also support for using Redis. * We have re-architected the structure of the Stack APIs to allow for more flexible routing. The motivating use cases are: - routing model A to ollama and model B to a remote provider like Together - routing shield A to local impl while shield B to a remote provider like Bedrock - routing a vector memory bank to Weaviate while routing a keyvalue memory bank to Redis * Support for provider specific parameters to be passed from the clients. A client can pass data using `x_llamastack_provider_data` parameter which can be type-checked and provided to the Adapter implementations.
This commit is contained in:
Ashwin Bharambe
GitHub
parent
8bf8c07eb3
commit
ec4fc800cc
130 changed files with 9701 additions and 11227 deletions
50
llama_stack/distribution/routers/__init__.py
Normal file
50
llama_stack/distribution/routers/__init__.py
Normal file
|
|
@ -0,0 +1,50 @@
|
|||
# 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, List, Tuple
|
||||
|
||||
from llama_stack.distribution.datatypes import * # noqa: F403
|
||||
|
||||
|
||||
async def get_routing_table_impl(
|
||||
api: Api,
|
||||
inner_impls: List[Tuple[str, Any]],
|
||||
routing_table_config: Dict[str, List[RoutableProviderConfig]],
|
||||
_deps,
|
||||
) -> Any:
|
||||
from .routing_tables import (
|
||||
MemoryBanksRoutingTable,
|
||||
ModelsRoutingTable,
|
||||
ShieldsRoutingTable,
|
||||
)
|
||||
|
||||
api_to_tables = {
|
||||
"memory_banks": MemoryBanksRoutingTable,
|
||||
"models": ModelsRoutingTable,
|
||||
"shields": ShieldsRoutingTable,
|
||||
}
|
||||
if api.value not in api_to_tables:
|
||||
raise ValueError(f"API {api.value} not found in router map")
|
||||
|
||||
impl = api_to_tables[api.value](inner_impls, routing_table_config)
|
||||
await impl.initialize()
|
||||
return impl
|
||||
|
||||
|
||||
async def get_auto_router_impl(api: Api, routing_table: RoutingTable, _deps) -> Any:
|
||||
from .routers import InferenceRouter, MemoryRouter, SafetyRouter
|
||||
|
||||
api_to_routers = {
|
||||
"memory": MemoryRouter,
|
||||
"inference": InferenceRouter,
|
||||
"safety": SafetyRouter,
|
||||
}
|
||||
if api.value not in api_to_routers:
|
||||
raise ValueError(f"API {api.value} not found in router map")
|
||||
|
||||
impl = api_to_routers[api.value](routing_table)
|
||||
await impl.initialize()
|
||||
return impl
|
||||
169
llama_stack/distribution/routers/routers.py
Normal file
169
llama_stack/distribution/routers/routers.py
Normal file
|
|
@ -0,0 +1,169 @@
|
|||
# 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, AsyncGenerator, Dict, List
|
||||
|
||||
from llama_stack.distribution.datatypes import RoutingTable
|
||||
|
||||
from llama_stack.apis.memory import * # noqa: F403
|
||||
from llama_stack.apis.inference import * # noqa: F403
|
||||
from llama_stack.apis.safety import * # noqa: F403
|
||||
|
||||
|
||||
class MemoryRouter(Memory):
|
||||
"""Routes to an provider based on the memory bank type"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
routing_table: RoutingTable,
|
||||
) -> None:
|
||||
self.routing_table = routing_table
|
||||
self.bank_id_to_type = {}
|
||||
|
||||
async def initialize(self) -> None:
|
||||
pass
|
||||
|
||||
async def shutdown(self) -> None:
|
||||
pass
|
||||
|
||||
def get_provider_from_bank_id(self, bank_id: str) -> Any:
|
||||
bank_type = self.bank_id_to_type.get(bank_id)
|
||||
if not bank_type:
|
||||
raise ValueError(f"Could not find bank type for {bank_id}")
|
||||
|
||||
provider = self.routing_table.get_provider_impl(bank_type)
|
||||
if not provider:
|
||||
raise ValueError(f"Could not find provider for {bank_type}")
|
||||
return provider
|
||||
|
||||
async def create_memory_bank(
|
||||
self,
|
||||
name: str,
|
||||
config: MemoryBankConfig,
|
||||
url: Optional[URL] = None,
|
||||
) -> MemoryBank:
|
||||
bank_type = config.type
|
||||
bank = await self.routing_table.get_provider_impl(bank_type).create_memory_bank(
|
||||
name, config, url
|
||||
)
|
||||
self.bank_id_to_type[bank.bank_id] = bank_type
|
||||
return bank
|
||||
|
||||
async def get_memory_bank(self, bank_id: str) -> Optional[MemoryBank]:
|
||||
provider = self.get_provider_from_bank_id(bank_id)
|
||||
return await provider.get_memory_bank(bank_id)
|
||||
|
||||
async def insert_documents(
|
||||
self,
|
||||
bank_id: str,
|
||||
documents: List[MemoryBankDocument],
|
||||
ttl_seconds: Optional[int] = None,
|
||||
) -> None:
|
||||
return await self.get_provider_from_bank_id(bank_id).insert_documents(
|
||||
bank_id, documents, ttl_seconds
|
||||
)
|
||||
|
||||
async def query_documents(
|
||||
self,
|
||||
bank_id: str,
|
||||
query: InterleavedTextMedia,
|
||||
params: Optional[Dict[str, Any]] = None,
|
||||
) -> QueryDocumentsResponse:
|
||||
return await self.get_provider_from_bank_id(bank_id).query_documents(
|
||||
bank_id, query, params
|
||||
)
|
||||
|
||||
|
||||
class InferenceRouter(Inference):
|
||||
"""Routes to an provider based on the model"""
|
||||
|
||||
def __init__(
|
||||
self,
|
||||
routing_table: RoutingTable,
|
||||
) -> None:
|
||||
self.routing_table = routing_table
|
||||
|
||||
async def initialize(self) -> None:
|
||||
pass
|
||||
|
||||
async def shutdown(self) -> None:
|
||||
pass
|
||||
|
||||
async def chat_completion(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[Message],
|
||||
sampling_params: Optional[SamplingParams] = SamplingParams(),
|
||||
tools: Optional[List[ToolDefinition]] = None,
|
||||
tool_choice: Optional[ToolChoice] = ToolChoice.auto,
|
||||
tool_prompt_format: Optional[ToolPromptFormat] = ToolPromptFormat.json,
|
||||
stream: Optional[bool] = False,
|
||||
logprobs: Optional[LogProbConfig] = None,
|
||||
) -> AsyncGenerator:
|
||||
# TODO: we need to fix streaming response to align provider implementations with Protocol.
|
||||
async for chunk in self.routing_table.get_provider_impl(model).chat_completion(
|
||||
model=model,
|
||||
messages=messages,
|
||||
sampling_params=sampling_params,
|
||||
tools=tools or [],
|
||||
tool_choice=tool_choice,
|
||||
tool_prompt_format=tool_prompt_format,
|
||||
stream=stream,
|
||||
logprobs=logprobs,
|
||||
):
|
||||
yield chunk
|
||||
|
||||
async def completion(
|
||||
self,
|
||||
model: str,
|
||||
content: InterleavedTextMedia,
|
||||
sampling_params: Optional[SamplingParams] = SamplingParams(),
|
||||
stream: Optional[bool] = False,
|
||||
logprobs: Optional[LogProbConfig] = None,
|
||||
) -> Union[CompletionResponse, CompletionResponseStreamChunk]:
|
||||
return await self.routing_table.get_provider_impl(model).completion(
|
||||
model=model,
|
||||
content=content,
|
||||
sampling_params=sampling_params,
|
||||
stream=stream,
|
||||
logprobs=logprobs,
|
||||
)
|
||||
|
||||
async def embeddings(
|
||||
self,
|
||||
model: str,
|
||||
contents: List[InterleavedTextMedia],
|
||||
) -> EmbeddingsResponse:
|
||||
return await self.routing_table.get_provider_impl(model).embeddings(
|
||||
model=model,
|
||||
contents=contents,
|
||||
)
|
||||
|
||||
|
||||
class SafetyRouter(Safety):
|
||||
def __init__(
|
||||
self,
|
||||
routing_table: RoutingTable,
|
||||
) -> None:
|
||||
self.routing_table = routing_table
|
||||
|
||||
async def initialize(self) -> None:
|
||||
pass
|
||||
|
||||
async def shutdown(self) -> None:
|
||||
pass
|
||||
|
||||
async def run_shield(
|
||||
self,
|
||||
shield_type: str,
|
||||
messages: List[Message],
|
||||
params: Dict[str, Any] = None,
|
||||
) -> RunShieldResponse:
|
||||
return await self.routing_table.get_provider_impl(shield_type).run_shield(
|
||||
shield_type=shield_type,
|
||||
messages=messages,
|
||||
params=params,
|
||||
)
|
||||
116
llama_stack/distribution/routers/routing_tables.py
Normal file
116
llama_stack/distribution/routers/routing_tables.py
Normal file
|
|
@ -0,0 +1,116 @@
|
|||
# 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, List, Optional, Tuple
|
||||
|
||||
from llama_models.sku_list import resolve_model
|
||||
from llama_models.llama3.api.datatypes import * # noqa: F403
|
||||
|
||||
from llama_stack.apis.models import * # noqa: F403
|
||||
from llama_stack.apis.shields import * # noqa: F403
|
||||
from llama_stack.apis.memory_banks import * # noqa: F403
|
||||
|
||||
from llama_stack.distribution.datatypes import * # noqa: F403
|
||||
|
||||
|
||||
class CommonRoutingTableImpl(RoutingTable):
|
||||
def __init__(
|
||||
self,
|
||||
inner_impls: List[Tuple[str, Any]],
|
||||
routing_table_config: Dict[str, List[RoutableProviderConfig]],
|
||||
) -> None:
|
||||
self.providers = {k: v for k, v in inner_impls}
|
||||
self.routing_keys = list(self.providers.keys())
|
||||
self.routing_table_config = routing_table_config
|
||||
|
||||
async def initialize(self) -> None:
|
||||
pass
|
||||
|
||||
async def shutdown(self) -> None:
|
||||
for p in self.providers.values():
|
||||
await p.shutdown()
|
||||
|
||||
def get_provider_impl(self, routing_key: str) -> Optional[Any]:
|
||||
return self.providers.get(routing_key)
|
||||
|
||||
def get_routing_keys(self) -> List[str]:
|
||||
return self.routing_keys
|
||||
|
||||
def get_provider_config(self, routing_key: str) -> Optional[GenericProviderConfig]:
|
||||
for entry in self.routing_table_config:
|
||||
if entry.routing_key == routing_key:
|
||||
return entry
|
||||
return None
|
||||
|
||||
|
||||
class ModelsRoutingTable(CommonRoutingTableImpl, Models):
|
||||
|
||||
async def list_models(self) -> List[ModelServingSpec]:
|
||||
specs = []
|
||||
for entry in self.routing_table_config:
|
||||
model_id = entry.routing_key
|
||||
specs.append(
|
||||
ModelServingSpec(
|
||||
llama_model=resolve_model(model_id),
|
||||
provider_config=entry,
|
||||
)
|
||||
)
|
||||
return specs
|
||||
|
||||
async def get_model(self, core_model_id: str) -> Optional[ModelServingSpec]:
|
||||
for entry in self.routing_table_config:
|
||||
if entry.routing_key == core_model_id:
|
||||
return ModelServingSpec(
|
||||
llama_model=resolve_model(core_model_id),
|
||||
provider_config=entry,
|
||||
)
|
||||
return None
|
||||
|
||||
|
||||
class ShieldsRoutingTable(CommonRoutingTableImpl, Shields):
|
||||
|
||||
async def list_shields(self) -> List[ShieldSpec]:
|
||||
specs = []
|
||||
for entry in self.routing_table_config:
|
||||
specs.append(
|
||||
ShieldSpec(
|
||||
shield_type=entry.routing_key,
|
||||
provider_config=entry,
|
||||
)
|
||||
)
|
||||
return specs
|
||||
|
||||
async def get_shield(self, shield_type: str) -> Optional[ShieldSpec]:
|
||||
for entry in self.routing_table_config:
|
||||
if entry.routing_key == shield_type:
|
||||
return ShieldSpec(
|
||||
shield_type=entry.routing_key,
|
||||
provider_config=entry,
|
||||
)
|
||||
return None
|
||||
|
||||
|
||||
class MemoryBanksRoutingTable(CommonRoutingTableImpl, MemoryBanks):
|
||||
|
||||
async def list_available_memory_banks(self) -> List[MemoryBankSpec]:
|
||||
specs = []
|
||||
for entry in self.routing_table_config:
|
||||
specs.append(
|
||||
MemoryBankSpec(
|
||||
bank_type=entry.routing_key,
|
||||
provider_config=entry,
|
||||
)
|
||||
)
|
||||
return specs
|
||||
|
||||
async def get_serving_memory_bank(self, bank_type: str) -> Optional[MemoryBankSpec]:
|
||||
for entry in self.routing_table_config:
|
||||
if entry.routing_key == bank_type:
|
||||
return MemoryBankSpec(
|
||||
bank_type=entry.routing_key,
|
||||
provider_config=entry,
|
||||
)
|
||||
return None
|
||||
Loading…
Add table
Add a link
Reference in a new issue