NVIDIA / NVIDIA/TensorRT-LLM

[Usage]: Optimal Config for NVFP4 - Qwen3 Next

Open
#10,159 2 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

General perf question
Dominant language
Python
Stars
14.7k
Forks
2.8k
Avg merge
2d 23h
Merged PRs (30d)
489

Description

System Info

Serving on B200 and getting mediocre results with the stock config

max_batch_size: 16
max_num_tokens: 4096
tensor_parallel_size: 4
moe_expert_parallel_size: 4
trust_remote_code: true
enable_attention_dp: false
cuda_graph_config:
  enable_padding: true
  max_batch_size: 720
moe_config:
    backend: TRTLLM
stream_interval: 20
num_postprocess_workers: 4
kv_cache_config:
  enable_block_reuse: false
  free_gpu_memory_fraction: 0.6

are there any weird ways to improve nvfp4 performance hre?

How would you like to use TensorRT-LLM

serving qwen3Next in fp4

Before submitting a new issue...
  • Make sure you already searched for relevant issues, and checked the documentation and examples for answers to frequently asked questions.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Use the supplied serving configuration as the baseline and consult the TensorRT-LLM documentation and examples for Qwen3 Next and NVFP4. Measure B200 serving performance after each configuration change; done means identifying a configuration that improves the reported results and documenting the settings used.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.