Hugging Face Journal Club: AsyncOPD and How Stale Can On-Policy Distillation Be?

From HuggingFace

2026-07-21 15:08 · Open original ↗

The Hugging Face post-training team discusses how on-policy distillation (OPD) can be made fully asynchronous to improve the throughput of training by 2-3x Paper: https://huggingface.co/papers/2606.24143

The Hugging Face post-training team discusses how on-policy distillation (OPD) can be made fully asynchronous to improve the throughput of training by 2-3x Paper: https://huggingface.co/papers/2606.24143