sgl-project/sglang-jax

[Perf] test Qwen/Qwen3-235B-A22B for RL

Open

#771 opened on Feb 6, 2026

View on GitHub
 (0 comments) (0 reactions) (1 assignee)Python (276 stars) (101 forks)auto 404
help wantedperformance

Description

  • model: Qwen/Qwen3-235B-A22B
  • chip: v6e
  • input: [4, 64, 256, 512, 1024, 2048, 4096]
  • output: [1, 4, 1024, 4096, 8192, 16384, 24576]
  • batch_size: [8,16,32,64,128,256,512]
  • mode: McJAX
  • hardware: tpu-v6e-32

Contributor guide

Tech stack
python
Domain
machine learning
Issue type
test
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
fresh
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
clear
Prerequisites
PythonJAXMachine Learning basics
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
20
Research direction
Set up the environment to run the model Qwen/Qwen3-235B A22B on TPU v6e with JAX, execute the given input shapes and batch sizes, measure latency and throughput, and compare against expected performance baselines.