Skip to content
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

Jump to bottom

[FEATURE]: [PyTorch] per-channel FP8 quantization #5873

Open
BurkeHulk opened this issue Jul 1, 2024 · 0 comments · Fixed by #5885
Open

[FEATURE]: [PyTorch] per-channel FP8 quantization #5873

BurkeHulk opened this issue Jul 1, 2024 · 0 comments · Fixed by #5885
Assignees
@BurkeHulk

Comments

@BurkeHulk
Copy link

BurkeHulk commented Jul 1, 2024 •
edited
Loading

Implement per-channel scaling (in PyTorch) for FP8 quantization.
Support PyTorch native FP8 formats.
Refer to:
https://pytorch.org/docs/stable/tensors.html#id7
https://arxiv.org/pdf/2209.05433
@BurkeHulk BurkeHulk changed the title [PyTorch] per-channel FP8 quantization [FEATURE] FP8 quantization in communication Jul 4, 2024
@BurkeHulk BurkeHulk mentioned this issue Jul 4, 2024
Merged
4 tasks
@BurkeHulk BurkeHulk reopened this Jul 4, 2024
@BurkeHulk BurkeHulk self-assigned this Jul 4, 2024
@BurkeHulk BurkeHulk changed the title [FEATURE] FP8 quantization in communication [FEATURE]: [PyTorch] per-channel FP8 quantization Jul 4, 2024
@ver217 ver217 linked a pull request Jul 16, 2024 that will close this issue
[feature]: support FP8 communication in pipeline parallelism #5885
Merged
4 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@BurkeHulk BurkeHulk

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

[feature]: support FP8 communication in pipeline parallelism BurkeHulk/ColossalAI
1 participant
@BurkeHulk