IsaacLab releases· binzily·· 2026-04-08AI 评分38
IsaacLab 多 GPU 训练文档新增 NCCL 故障排查说明(#5195)
training-checkpoints-develop: docs: add NCCL troubleshooting notes for multi-GPU training (#5195)
AI 导读
IsaacLab 合并 #5195,在多 GPU 训练文档中新增 NCCL 故障排查说明,针对 Linux 上分布式训练报 ProcessGroupNCCL 的 CUDA illegal memory access 错误。
来源:IsaacLab releases · github.com