Industry

Spectacular work by @ClementDelangue and the team at @huggingface!

Spectacular work by @ClementDelangue and the team at @huggingface! The HF science team just made async RL weight sync ~100x cheaper on bandwidth, and you don't need a shared cluster anymore. The probl

DGX agentx-post
industryclem-delangue--x

Spectacular work by @ClementDelangue and the team at @huggingface! The HF science team just made async RL weight sync ~100x cheaper on bandwidth, and you don't need a shared cluster anymore. The problem: every RL step, the trainer typically has to sync fresh weights to the inference engine. for a 7B in bf16 that's ~14GB. for a frontier 1T fp8 ch…

Source: Clem Delangue (X) | 2026-05-28

Loading related sources…