4 | @humansand
Production Blackwell-Native MXFP8 & NVFP4 RL with SGLang and Miles
Ziang Li shares lessons on trainer-sampler consistency, gradient stability, and precision control when bringing low-precision RL to production.
Blackwell-native MXFP8 and NVFP4 RL with SGLang and Miles lessons
AIZiang Li from humansand shares lessons on trainer-sampler consistency, gradient stability, and precision control for bringing low-precision reinforcement learning with MXFP8 and NVFP4 on Blackwell into production. The post, shared by LMSYS Org, covers production RL built on SGLang and Miles.
@lmsysorg
Part of a thread · earlier post