forked from phoenix-oss/llama-stack-mirror
6bb57e72a7
This PR makes several core changes to the developer experience surrounding Llama Stack. Background: PR #92 introduced the notion of "routing" to the Llama Stack. It introduces three object types: (1) models, (2) shields and (3) memory banks. Each of these objects can be associated with a distinct provider. So you can get model A to be inferenced locally while model B, C can be inference remotely (e.g.) However, this had a few drawbacks: you could not address the provider instances -- i.e., if you configured "meta-reference" with a given model, you could not assign an identifier to this instance which you could re-use later. the above meant that you could not register a "routing_key" (e.g. model) dynamically and say "please use this existing provider I have already configured" for a new model. the terms "routing_table" and "routing_key" were exposed directly to the user. in my view, this is way too much overhead for a new user (which almost everyone is.) people come to the stack wanting to do ML and encounter a completely unexpected term. What this PR does: This PR structures the run config with only a single prominent key: - providers Providers are instances of configured provider types. Here's an example which shows two instances of the remote::tgi provider which are serving two different models. providers: inference: - provider_id: foo provider_type: remote::tgi config: { ... } - provider_id: bar provider_type: remote::tgi config: { ... } Secondly, the PR adds dynamic registration of { models | shields | memory_banks } to the API surface. The distribution still acts like a "routing table" (as previously) except that it asks the backing providers for a listing of these objects. For example it asks a TGI or Ollama inference adapter what models it is serving. Only the models that are being actually served can be requested by the user for inference. Otherwise, the Stack server will throw an error. When dynamically registering these objects, you can use the provider IDs shown above. Info about providers can be obtained using the Api.inspect set of endpoints (/providers, /routes, etc.) The above examples shows the correspondence between inference providers and models registry items. Things work similarly for the safety <=> shields and memory <=> memory_banks pairs. Registry: This PR also makes it so that Providers need to implement additional methods for registering and listing objects. For example, each Inference provider is now expected to implement the ModelsProtocolPrivate protocol (naming is not great!) which consists of two methods register_model list_models The goal is to inform the provider that a certain model needs to be supported so the provider can make any relevant backend changes if needed (or throw an error if the model cannot be supported.) There are many other cleanups included some of which are detailed in a follow-up comment.
217 lines
6 KiB
Python
217 lines
6 KiB
Python
# 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 enum import Enum
|
|
|
|
from typing import List, Literal, Optional, Protocol, runtime_checkable, Union
|
|
|
|
from llama_models.schema_utils import json_schema_type, webmethod
|
|
|
|
from pydantic import BaseModel, Field
|
|
from typing_extensions import Annotated
|
|
|
|
from llama_models.llama3.api.datatypes import * # noqa: F403
|
|
from llama_stack.apis.models import * # noqa: F403
|
|
|
|
|
|
class LogProbConfig(BaseModel):
|
|
top_k: Optional[int] = 0
|
|
|
|
|
|
@json_schema_type
|
|
class QuantizationType(Enum):
|
|
bf16 = "bf16"
|
|
fp8 = "fp8"
|
|
|
|
|
|
@json_schema_type
|
|
class Fp8QuantizationConfig(BaseModel):
|
|
type: Literal[QuantizationType.fp8.value] = QuantizationType.fp8.value
|
|
|
|
|
|
@json_schema_type
|
|
class Bf16QuantizationConfig(BaseModel):
|
|
type: Literal[QuantizationType.bf16.value] = QuantizationType.bf16.value
|
|
|
|
|
|
QuantizationConfig = Annotated[
|
|
Union[Bf16QuantizationConfig, Fp8QuantizationConfig],
|
|
Field(discriminator="type"),
|
|
]
|
|
|
|
|
|
@json_schema_type
|
|
class ChatCompletionResponseEventType(Enum):
|
|
start = "start"
|
|
complete = "complete"
|
|
progress = "progress"
|
|
|
|
|
|
@json_schema_type
|
|
class ToolCallParseStatus(Enum):
|
|
started = "started"
|
|
in_progress = "in_progress"
|
|
failure = "failure"
|
|
success = "success"
|
|
|
|
|
|
@json_schema_type
|
|
class ToolCallDelta(BaseModel):
|
|
content: Union[str, ToolCall]
|
|
parse_status: ToolCallParseStatus
|
|
|
|
|
|
@json_schema_type
|
|
class ChatCompletionResponseEvent(BaseModel):
|
|
"""Chat completion response event."""
|
|
|
|
event_type: ChatCompletionResponseEventType
|
|
delta: Union[str, ToolCallDelta]
|
|
logprobs: Optional[List[TokenLogProbs]] = None
|
|
stop_reason: Optional[StopReason] = None
|
|
|
|
|
|
@json_schema_type
|
|
class CompletionRequest(BaseModel):
|
|
model: str
|
|
content: InterleavedTextMedia
|
|
sampling_params: Optional[SamplingParams] = SamplingParams()
|
|
|
|
stream: Optional[bool] = False
|
|
logprobs: Optional[LogProbConfig] = None
|
|
|
|
|
|
@json_schema_type
|
|
class CompletionResponse(BaseModel):
|
|
"""Completion response."""
|
|
|
|
completion_message: CompletionMessage
|
|
logprobs: Optional[List[TokenLogProbs]] = None
|
|
|
|
|
|
@json_schema_type
|
|
class CompletionResponseStreamChunk(BaseModel):
|
|
"""streamed completion response."""
|
|
|
|
delta: str
|
|
stop_reason: Optional[StopReason] = None
|
|
logprobs: Optional[List[TokenLogProbs]] = None
|
|
|
|
|
|
@json_schema_type
|
|
class BatchCompletionRequest(BaseModel):
|
|
model: str
|
|
content_batch: List[InterleavedTextMedia]
|
|
sampling_params: Optional[SamplingParams] = SamplingParams()
|
|
logprobs: Optional[LogProbConfig] = None
|
|
|
|
|
|
@json_schema_type
|
|
class BatchCompletionResponse(BaseModel):
|
|
"""Batch completion response."""
|
|
|
|
completion_message_batch: List[CompletionMessage]
|
|
|
|
|
|
@json_schema_type
|
|
class ChatCompletionRequest(BaseModel):
|
|
model: str
|
|
messages: List[Message]
|
|
sampling_params: Optional[SamplingParams] = SamplingParams()
|
|
|
|
# zero-shot tool definitions as input to the model
|
|
tools: Optional[List[ToolDefinition]] = Field(default_factory=list)
|
|
tool_choice: Optional[ToolChoice] = Field(default=ToolChoice.auto)
|
|
tool_prompt_format: Optional[ToolPromptFormat] = Field(
|
|
default=ToolPromptFormat.json
|
|
)
|
|
|
|
stream: Optional[bool] = False
|
|
logprobs: Optional[LogProbConfig] = None
|
|
|
|
|
|
@json_schema_type
|
|
class ChatCompletionResponseStreamChunk(BaseModel):
|
|
"""SSE-stream of these events."""
|
|
|
|
event: ChatCompletionResponseEvent
|
|
|
|
|
|
@json_schema_type
|
|
class ChatCompletionResponse(BaseModel):
|
|
"""Chat completion response."""
|
|
|
|
completion_message: CompletionMessage
|
|
logprobs: Optional[List[TokenLogProbs]] = None
|
|
|
|
|
|
@json_schema_type
|
|
class BatchChatCompletionRequest(BaseModel):
|
|
model: str
|
|
messages_batch: List[List[Message]]
|
|
sampling_params: Optional[SamplingParams] = SamplingParams()
|
|
|
|
# zero-shot tool definitions as input to the model
|
|
tools: Optional[List[ToolDefinition]] = Field(default_factory=list)
|
|
tool_choice: Optional[ToolChoice] = Field(default=ToolChoice.auto)
|
|
tool_prompt_format: Optional[ToolPromptFormat] = Field(
|
|
default=ToolPromptFormat.json
|
|
)
|
|
logprobs: Optional[LogProbConfig] = None
|
|
|
|
|
|
@json_schema_type
|
|
class BatchChatCompletionResponse(BaseModel):
|
|
completion_message_batch: List[CompletionMessage]
|
|
|
|
|
|
@json_schema_type
|
|
class EmbeddingsResponse(BaseModel):
|
|
embeddings: List[List[float]]
|
|
|
|
|
|
class ModelStore(Protocol):
|
|
def get_model(self, identifier: str) -> ModelDef: ...
|
|
|
|
|
|
@runtime_checkable
|
|
class Inference(Protocol):
|
|
model_store: ModelStore
|
|
|
|
# This method is not `async def` because it can result in either an
|
|
# `AsyncGenerator` or a `CompletionResponse` depending on the value of `stream`.
|
|
@webmethod(route="/inference/completion")
|
|
def completion(
|
|
self,
|
|
model: str,
|
|
content: InterleavedTextMedia,
|
|
sampling_params: Optional[SamplingParams] = SamplingParams(),
|
|
stream: Optional[bool] = False,
|
|
logprobs: Optional[LogProbConfig] = None,
|
|
) -> Union[CompletionResponse, CompletionResponseStreamChunk]: ...
|
|
|
|
# This method is not `async def` because it can result in either an
|
|
# `AsyncGenerator` or a `ChatCompletionResponse` depending on the value of `stream`.
|
|
@webmethod(route="/inference/chat_completion")
|
|
def chat_completion(
|
|
self,
|
|
model: str,
|
|
messages: List[Message],
|
|
sampling_params: Optional[SamplingParams] = SamplingParams(),
|
|
# zero-shot tool definitions as input to the model
|
|
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,
|
|
) -> Union[ChatCompletionResponse, ChatCompletionResponseStreamChunk]: ...
|
|
|
|
@webmethod(route="/inference/embeddings")
|
|
async def embeddings(
|
|
self,
|
|
model: str,
|
|
contents: List[InterleavedTextMedia],
|
|
) -> EmbeddingsResponse: ...
|