yandex/YaFSDP
YaFSDP: Yet another Fully Sharded Data Parallel
135 best points · 16 comments · 2 Hacker News posts · 986 GitHub stars · View on GitHub
Hacker News discussions
- YaFSDP: a sharded data parallelism framework, faster for pre-training LLMs (135 points, 16 comments, by wiradikusuma, 2024-06-18)
- Yandex Unveils YaFSDP for 26% Faster LLM Training (12 points, 0 comments, by annaerma, 2024-06-11)