Repository navigation
Conversation
…h unaligned rows For SAME_BOTH_DIMS, group_quantize only checked that the stacked rows were a multiple of 128. When the per-tensor rows were not (e.g. two 192 x 256 tensors), the last chunk of each tensor ran into the next one: the last tensor's scales were written past the end of the scale buffers, and the dbias partials were mixed between tensors and summed from the wrong workspace rows. Require the rows of each tensor to be a multiple of 128 on the host, where they are known for this representation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: Kaining Zhong <kainingz@nvidia.com>
[Medium impact] The PR affects validation for same-shaped grouped MXFP8 quantization. The change appears safe to merge, with a non-blocking request to test the new rejection. Findings
SummaryAdds a host-side check requiring each
Reviews (1) · Last reviewed commit: "Merge branch 'main' into fix_grouped_mxf..." · Reviewed by Greptile |
| if (shape_rep == ShapeRepresentation::SAME_BOTH_DIMS) { | ||
| NVTE_CHECK( | ||
| first_logical_dim % num_tensors == 0 && (first_logical_dim / num_tensors) % tile_dim_y == 0, | ||
| "First dimension of each tensor in a group must be divisible by ", tile_dim_y, ", got ", | ||
| first_logical_dim, " rows split across ", num_tensors, " tensors."); |
There was a problem hiding this comment.
The existing grouped MXFP8 test skips members whose row count is not divisible by 128, so adding an invalid shape to its input list would not exercise this check. Add a standalone rejection test for two 64-row tensors: their total of 128 rows passes the old check but must fail the new one. Keep an aligned success case alongside it. This would catch future changes that accidentally remove the protection.
Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
|
/te-ci |
Description
Fixes #3911
Type of change
Changes
Please list the changes introduced in this PR:
Checklist: