在Windows下复现pyskl训练报错：RuntimeError: Distributed package doesn’t have NCCL

Saturn-

已于 2023-12-27 18:23:44 修改

阅读量690

点赞数 9

文章标签：深度学习

于 2023-11-24 10:03:56 首次发布

本文链接：https://blog.csdn.net/weixin_45600637/article/details/134591787

版权

pyskl使用以下命令训练时报错:

bash tools/dist_train.sh configs/stgcn++/stgcn++_ntu120_xsub_hrnet/j.py 1 --validate --test-last --test-best

1. 定位：

C:\Users\lib\site-packages\torch\distributed\distributed_c10d.py

2. 找到:

prefix_store = PrefixStore(group_name, store)

3. 修改前：

prefix_store = PrefixStore(group_name, store)
if backend == Backend.GLOO:
    pg = ProcessGroupGloo(
        prefix_store,
        rank,
        world_size,
        timeout=timeout)
    _pg_map[pg] = (Backend.GLOO, store)
    _pg_names[pg] = group_name
elif backend == Backend.NCCL:
    if not is_nccl_available():
        raise RuntimeError("Distributed package doesn't have NCCL "
                           "built in")

4. 修改后：

prefix_store = PrefixStore(group_name, store)
backend = "gloo"
if backend == Backend.GLOO:
    pg = ProcessGroupGloo(
        prefix_store,
        rank,
        world_size,
        timeout=timeout)
    _pg_map[pg] = (Backend.GLOO, store)
    _pg_names[pg] = group_name
elif backend == Backend.NCCL:
    if not is_nccl_available():
        raise RuntimeError("Distributed package doesn't have NCCL "
                           "built in")