Repository navigation
Conversation
🏷️ CI GuideRuns automatically on every PR:
Extended tests (opt-in via labels):
|
645cf33 to
c636f41
Compare
aa917e2 to
1b55b0b
Compare
|
@XiaobingSuper Could you help review this PR? I'm not sure whether the change will affect GPT-OSS |
For Swiglu MXFP4 bf16-activation CK-Tile dispatch, use n_pad_zeros=0 so padded SEPARATED w13 layouts are handled correctly while keeping the fast bf16 path. Scope the behavior to swiglu_mxfp4_bf16_cktile only; preserve the legacy n_pad_zeros formula for generic ksplit>1 paths to avoid impacting other models. Add regression guards for both target and non-target paths in op_tests/test_moe_2stage_npad_guard.py.
1b55b0b to
96254d1
Compare
|
hello, I've tested the accuracy, and there are no issues of MiniMax M3 MXFP4 on ATOM with TP=8 using the current code(origin/main). server lm-eval We're using MiniMax-M3-MXFP4, not MiniMax-M3-MXFP4-AttnFP8.
|
|
The issue was fixed indeed.
I will close this PR |

For Swiglu MXFP4 bf16-activation CK-Tile dispatch, use n_pad_zeros=0 so padded SEPARATED w13 layouts are handled correctly while keeping the fast bf16 path.
Scope the behavior to swiglu_mxfp4_bf16_cktile only; preserve the legacy n_pad_zeros formula for generic ksplit>1 paths to avoid impacting other models.
Add regression guards for both target and non-target paths in op_tests/test_moe_2stage_npad_guard.py.
Motivation
Properly run MiniMax M3 MXFP4 on ATOM with TP=8
Test Result
Tested with ATOM with TP=8:
lm-eval
Results, for reference with TP=1 gsm8k gives ~0.9424:
origin/main(no fix)Submission Checklist