00:00
2026-09-10
huggingface.co
ai-research
Async GRPO with LoRA across HF Jobs: a bucket, a proxy, and no NCCL
TRL v1.14 shipped LoRA support in its AsyncGRPOTrainer via PR #7017, letting the asynchronous trainer train a rank-1 LoRA adapter and sync only that adapter to vLLM instead of the full model. A Huggin…