Excited to share our work on large-scale sharded weight transfer! We’ve implemented a native sharded weight transfer engine in vLLM and SkyRL using Ray Direct Transfer (RDT) + NIXL. We’re able to achieve weight transfer for Kimi K2 (1T params) in BF16 in 7.53 seconds on 48…
This weight transfer engine is native to vLLM. Any Ray-based trainer can adopt it with a single WeightSource iterator. Great work here! 🔗
This weight transfer engine is native to vLLM. Any Ray-based trainer can adopt it with a single WeightSource iterator. Great work here! 🔗
15.6K
3 posts, first seen 1d ago
Excited to share our work on large-scale sharded weight transfer! We’ve implemented a native sharded weight transfer engine in vLLM and SkyRL using Ray Direct Transfer (RDT) + NIXL. We’re able to achieve weight transfer for Kimi K2 (1T params) in BF16 in 7.53 seconds on 48…
This weight transfer engine is native to vLLM. Any Ray-based trainer can adopt it with a single WeightSource iterator. Great work here! 🔗
This weight transfer engine is native to vLLM. Any Ray-based trainer can adopt it with a single WeightSource iterator. Great work here! 🔗