mirror of
https://github.com/meta-llama/llama-stack.git
synced 2025-06-28 02:53:30 +00:00
f34f22f8c7
# What does this PR do? This PR adds two methods to the Inference API: - `batch_completion` - `batch_chat_completion` The motivation is for evaluations targeting a local inference engine (like meta-reference or vllm) where batch APIs provide for a substantial amount of acceleration. Why did I not add this to `Api.batch_inference` though? That just resulted in a _lot_ more book-keeping given the structure of Llama Stack. Had I done that, I would have needed to create a notion of a "batch model" resource, setup routing based on that, etc. This does not sound ideal. So what's the future of the batch inference API? I am not sure. Maybe we can keep it for true _asynchronous_ execution. So you can submit requests, and it can return a Job instance, etc. ## Test Plan Run meta-reference-gpu using: ```bash export INFERENCE_MODEL=meta-llama/Llama-4-Scout-17B-16E-Instruct export INFERENCE_CHECKPOINT_DIR=../checkpoints/Llama-4-Scout-17B-16E-Instruct-20250331210000 export MODEL_PARALLEL_SIZE=4 export MAX_BATCH_SIZE=32 export MAX_SEQ_LEN=6144 LLAMA_MODELS_DEBUG=1 llama stack run meta-reference-gpu ``` Then run the batch inference test case.
55 lines
1.9 KiB
Python
55 lines
1.9 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 typing import List, Optional, Protocol, runtime_checkable
|
|
|
|
from llama_stack.apis.common.job_types import Job
|
|
from llama_stack.apis.inference import (
|
|
InterleavedContent,
|
|
LogProbConfig,
|
|
Message,
|
|
ResponseFormat,
|
|
SamplingParams,
|
|
ToolChoice,
|
|
ToolDefinition,
|
|
ToolPromptFormat,
|
|
)
|
|
from llama_stack.schema_utils import webmethod
|
|
|
|
|
|
@runtime_checkable
|
|
class BatchInference(Protocol):
|
|
"""Batch inference API for generating completions and chat completions.
|
|
|
|
This is an asynchronous API. If the request is successful, the response will be a job which can be polled for completion.
|
|
|
|
NOTE: This API is not yet implemented and is subject to change in concert with other asynchronous APIs
|
|
including (post-training, evals, etc).
|
|
"""
|
|
|
|
@webmethod(route="/batch-inference/completion", method="POST")
|
|
async def completion(
|
|
self,
|
|
model: str,
|
|
content_batch: List[InterleavedContent],
|
|
sampling_params: Optional[SamplingParams] = None,
|
|
response_format: Optional[ResponseFormat] = None,
|
|
logprobs: Optional[LogProbConfig] = None,
|
|
) -> Job: ...
|
|
|
|
@webmethod(route="/batch-inference/chat-completion", method="POST")
|
|
async def chat_completion(
|
|
self,
|
|
model: str,
|
|
messages_batch: List[List[Message]],
|
|
sampling_params: Optional[SamplingParams] = None,
|
|
# 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] = None,
|
|
response_format: Optional[ResponseFormat] = None,
|
|
logprobs: Optional[LogProbConfig] = None,
|
|
) -> Job: ...
|