Skip to content

feat: 122b has been tested - #22

Draft
fiskrt wants to merge 1 commit into
huawei-csl:mainfrom
fiskrt:feat-readme-122b-tested
Draft

feat: 122b has been tested#22
fiskrt wants to merge 1 commit into
huawei-csl:mainfrom
fiskrt:feat-readme-122b-tested

Conversation

@fiskrt

@fiskrt fiskrt commented Jun 12, 2026

Copy link
Copy Markdown
Contributor

We see a 37% increase in token/s throughput on the full GSM8k test set, while observing no loss in performance (pto is even better). Setup: Qwen 3.5 122B TP=4 and TP=8.

PTO backend
100%|██████████████████████████████████████████████████| 1319/1319 [19:22<00:00, 1.13it/s]
Accuracy: 0.961
Invalid: 0.000
Latency: 1162.949 s
Output throughput: 196.239 token/s

Triton backend
100%|██████████████████████████████████████████████████| 1319/1319 [26:11<00:00, 1.19s/it]
Accuracy: 0.954
Invalid: 0.000
Latency: 1571.015 s
Output throughput: 143.936 token/s

@zouzias zouzias added the GDN label Jul 26, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants