NVIDIA / NVIDIA/Megatron-LM

[feat] GTP+2D Mxfp8

Open
#6,363 0 comments 0 reactions 1 assignee Assigned to @dingqingy-nv View on GitHub
nemotron
Dominant language
Python
Stars
17.9k
Forks
4.5k
Avg merge
4d 3h
Merged PRs (30d)
272

Description

New 2D quant for mxfp8 support in TE: https://github.com/NVIDIA/TransformerEngine/pull/2634
- 2D quant is for only for weight tensors: `W, W^T = 2D quantized in blk 32x32`
- Activations are quantized using 1D scales (1x32 elements): `X, X^T = 1D quantized in blk 1x32`

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.