Skip to content

[PyTorch] Optionally derive FC1 dy NVFP4 amaxes from the dGLU epilogue #3755

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sraman-rgb wants to merge 3 commits into
base: main
Choose a base branch
from
+586 −8
Open

[PyTorch] Optionally derive FC1 dy NVFP4 amaxes from the dGLU epilogue #3755

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ffce0ef
[PyTorch] Optionally derive FC1 dy NVFP4 amaxes from the dGLU epilogue
sraman-rgb Oct 27, 2026
994ddd7
[PyTorch] Test the dGLU amax path in the fused grouped MLP backward
sraman-rgb Nov 10, 2026
bec1fe3
Merge remote-tracking branch 'upstream/main' into sraman/nvfp4-fc1-dy…
sraman-rgb Nov 13, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions docs/envvars.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -371,6 +371,12 @@ Kernel Configuration
:Default: ``0``
:Description: Enable row-scaled NVFP4 tensors for forward activation quantizers in the ``NVFP4BlockScaling`` recipe. When set to ``1`` (or when ``NVFP4BlockScaling(row_scaled_activation=True)`` is used), rowwise ``amax`` metadata is stored as one FP32 value per tensor row instead of a single scalar.

.. envvar:: NVTE_NVFP4_FC1_DY_AMAX_FROM_DGLU

:Type: ``int`` (0 or 1)
:Default: ``0``
:Description: In the fused grouped MLP backward with NVFP4 and a GLU activation, derive the per-expert amaxes for quantizing the FC1 grad output from the amax that the FC2 dGLU epilogue already computes, instead of a separate amax pass over the FC1 grad output. The rowwise amax is exact. With RHT, the columnwise (wgrad) amax is replaced by the upper bound ``4 * rowwise amax`` on the post-RHT amax (the 16-point RHT is normalized by 1/4), which is looser than the exact post-RHT amax (8-13x on DeepSeek-V3 gradients, i.e. up to about 3.7 binades of FP8 scale range left unused in the wgrad operand). The bound can never overflow; it only reduces the precision headroom of the columnwise quantization.

.. envvar:: NVTE_NVFP4_4OVER6

:Type: ``str`` (``none``, ``weights``, ``activations``, or ``all``)
Expand Down
Loading
Loading