Activity
Stars and mentions · Last 30 days, Oct 5 – Nov 3
Top repositories
stas00’s repositories ranked by new forks, as on trending.
Contributes to
Repositories outside stas00’s account where stas00 opened or reviewed pull requests in the past 365 days.
Who noticed stas00
Everyone who starred or forked stas00’s repositories since the archive began in 2011.
Notable accounts
Ordered by the stars their own repositories have received.
Biggest fans
Accounts that starred the most of these repositories.
Breakout weeks
Each repository’s best week of new stars, with the release published around it.
Achievements
Awards earned by the repositories stas00 owns. Select an award to embed its badge.
- Gold
- 11st place finishes
- Silver
- 02nd place finishes
- Bronze
- 03rd place finishes
- Awarded repositories
- 1Owned by stas00 now
Repositories
Ordered by gold, then silver, then bronze.
- ml-engineeringBest: 1st by stars, 1,679 stars100
Every award
Top-three finishes of these repositories in completed trending days and weeks, newest first.
A repository earns an award for finishing first, second or third in a completed trending day or week, ranked by new stars, new forks or pull request authors. Awards follow a repository through renames and transfers and count toward its current owner. Weeks run Monday to Sunday, UTC.
Mentions
FYI @nvidia has announced it downgraded the performance of NVLink 6 and HBM4 speed for its upcoming Rubin products - NVLInk 6 is down to 1.5TBps from 1.8TBps (unidirectional per GPU) - HBM4 is down to 19.2TBps from 22TBps github.com/stas00/ml-engineering/blob/master/netw… github.com/stas00/ml-engineering/blob/master/comp… There are various speculations online to why this has been done, but no official explanation I was able to find.
This is big news for those who need more gpu memory and have a ton of cpu memory @huggingface Transformers now allows you to offload activation checkpoints to CPU: github.com/stas00/ml-engineering/tree/master/trai… It needs more TLC to be faster but it's already amazing. If you want close to 0-overhead implementation, see ours in github.com/Snowflake-AI-Research/Arctic-Platform/… - it uses a separate CUDA stream w/ full overlap w/ compute and prefetch. p.s. you need transformers>=5.17
Introducing torch-dist-bench, very similar to nccl-tests but from pytorch perspective and only needs torch to run. Measures bandwidth and latency of all torch.dist collectives with tables and plots github.com/stas00/ml-engineering/tree/master/netw… this is a new incarnation of all-reduce-bench t.co
Lots of new gpu arch additions to the MAMF/MSMF table: github.com/stas00/ml-engineering/blob/master/comp… Thank you: - Jon @HotAisle - MI300X - @JordanNanos @SemiAnalysis_ - H100+MI325X - @a_e_roberts @huggingface - MI300X+350X - Billy Cao @SemiAnalysis_ H200 NVL t.co
My fellow NVFP4 nerds. If you somehow don’t know who Stas is, that was your first mistake. Did you know he gives away all his knowledge for literally free github.com/stas00/ml-engineering/tree/master