vllm-project/vllm

[Feature]: Batch Invariant Feature and Performance Optimization

Open

#27433 opened on Oct 23, 2025

View on GitHub
 (47 comments) (27 reactions) (3 assignees)Python (80,034 stars) (16,816 forks)batch import
feature requestgood first issue

Description

🚀 The feature, motivation and pitch

We have basically support Batch Invariant based on https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/

https://github.com/orgs/vllm-project/projects/29/views/1

But there are still some work to be done, so here is the issue to track the work

TODOs:

Nice to have:

Model coverage

https://docs.vllm.ai/en/latest/features/batch_invariance/#tested-models

🙋Help needed for validations of more models.

  1. Test a model using the script in https://github.com/vllm-project/vllm/tree/main/tests/v1/determinism
  2. Submit a PR updating the document

Contributor guide

Tech stack
pythonpytorch
Domain
aibackendperformance
Issue type
feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
5
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
over 1 week
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
Familiarity with vLLMCUDA programmingAttention mechanismsGPU optimization
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
15
Research direction
Look at the remaining help needed items specifically: FLASHINFER MLA support (see flashinfer issue #2107) and optimizing batch invariant performance (see PRs #29345, #40413, #40408). Understand the existing batch invariant framework via the linked blog post and the vLLM documentation at https://docs.vllm.ai/en/latest/features/batch invariance. Then, for FLASHINFER MLA, examine the flashinfer issue and the existing Triton MLA support to determine what implementation is needed. For performance optimization, review the merged PRs to identify remaining bottlenecks and propose further optimizations.