Thanks for the call out and glad to see that our blog inspired the broader community! We at @FireworksAI_HQ have been using delta updates to…

Thanks for the call out and glad to see that our blog inspired the broader community! We at @FireworksAI_HQ have been using delta updates to scale RL globally for about a year. Including Composer 2 & 2.5 training in 4 datacenters around the world 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) | 2026-05-29