Repository navigation
[bug] Qwen3-30B-A3B GB200 MXFP8 perf preset (full_iteration CUDA graphs + HybridEP) OOMs in first optimizer step on PyTorch >=2.14 / TE >=2.18 stacks #22344
Triggered via issue
October 12, 2026 20:59
janbernloehr
Status
Skipped
Total duration
11s
Artifacts
–
request-nvskills-ci.yml
on: issue_comment
request
/
request