Unleashing Computational Power: Ultimate Latency Optimization of Qwen3 and Qwen3-VL on AMD MI300X Series #313
+292
−0
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Author
wuhuikx
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
In recent months, the Qwen C-end Infrastructure Engineering Team and the AMD AIFW Team have collaborated to implement extreme latency optimization solutions for Qwen3 235B and Qwen3 VL 235B on the AMD MI300 series GPU platform, based on the SGLang framework.
Remarkable breakthroughs have been achieved in terms of performance, precision, and stability.
• For Qwen3 235B: Compared with the baseline, the Time to First Token (TTFT) has been improved by 1.67×, and the Time Per Output Token (TPOT) has been improved by 2.12×.
• For Qwen3 VL 235B: Compared with the baseline, the Time to First Token (TTFT) has been improved by 1.62×, and the Time Per Output Token (TPOT) has been improved by 1.90×.
This paper elaborates on the performance optimization techniques jointly explored and implemented by the two teams, with a core focus on achieving ultra-low-latency inference.