deepseek-ai/DualPipe
A bidirectional pipeline parallelism algorithm for computation-communication overlap in DeepSeek V3/R1 training.
180 best points · 20 comments · 1 Hacker News post · 3,030 GitHub stars · View on GitHub
Hacker News discussions
- DualPipe: Bidirectional pipeline parallelism algorithm (180 points, 20 comments, by mfiguiere, 2025-02-27)