[mxfp8] [docs] [BE] add MXFP8 usage documentation and benchmarks #2096
Conversation
danielvegamyhre
requested review from
fegin,
tianyu-l,
wconstab and
wwwjn
as code owners
meta-cla
bot
added
the
CLA Signed
danielvegamyhre
force-pushed
the
docmx
branch
5 times, most recently
from
57b3cf0 to
66c8771
Compare
tianyu-l
left a comment
tianyu-l
left a comment
There was a problem hiding this comment.
nit comments on picture
There was a problem hiding this comment.
maybe put in the existing assets/images folder
There was a problem hiding this comment.
The image looks very blurry, not sure why. If possible could you find a more clear version? Fine if not.
There was a problem hiding this comment.
moved the image to assets/images. as for the blurriness, it's the same image as in the torchao docs here, where i think it looks reasonably clear: https://github.com/pytorch/ao/tree/main/torchao/prototype/moe_training i'll see if we can get a clearer one though
danielvegamyhre
force-pushed
the
docmx
branch
2 times, most recently
from
3626393 to
7d68f7f
Compare
danielvegamyhre
force-pushed
the
docmx
branch
from
7d68f7f to
afbb18b
Compare
|
|
||
| **Hardware Requirements:** | ||
|
|
||
| MXFP8 training requires NVIDIA B200 (SM100) or newer GPUs. |
There was a problem hiding this comment.
seems like we can just say this once, not necessary to repeat multiple times
| - torchao version: `0.13.0+gite4e681be` | ||
| - torchtitan commit: `6fc499f6f5b32151a799188be2208cfb09faed30` | ||
|
|
||
| *Source: [TorchAO MX Formats Benchmarks](https://github.com/pytorch/ao/tree/main/torchao/prototype/mx_formats#training-e2e-benchmarks-on-nvidia-b200)* |
There was a problem hiding this comment.
nit: just link to the benchmarks instead of copy-paste, so that it can be updated in just one place
3 participants
Fixes #1998