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.

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

adding throughtput numbers and tools for getting numbers #53

Merged
merged 5 commits into from
Dec 21, 2022

Conversation

vchiley
Copy link
Contributor

@vchiley vchiley commented Dec 20, 2022

Adding Mosaic_GPT's throughput / MFU numbers.
Includes scripts using mcli's python api for launching a bunch of runs / configs

Copy link
Contributor

@abhi-mosaic abhi-mosaic left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks sweet! Approving for now as I think it would be good to get this merged quickly and update on the go. After torch 1.13 / composer 0.12 / Triton kernels we'll have to come back to this anyways..

llm/throughput/README.md Outdated Show resolved Hide resolved
llm/throughput/README.md Outdated Show resolved Hide resolved
llm/throughput/README.md Show resolved Hide resolved
llm/throughput/README.md Show resolved Hide resolved
llm/throughput/run_all_configs.py Outdated Show resolved Hide resolved
llm/throughput/run_all_configs.py Outdated Show resolved Hide resolved
@abhi-mosaic
Copy link
Contributor

I also have a feeling that we may be underselling the 1B and 3B models on 80GB cards, because they should work without activation checkpointing... but upgrading those #s can be a followup PR and I'll make a JIRA ticket to track.

@vchiley vchiley merged commit 22d724c into mosaicml:main Dec 21, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants