fix tensor data update for gemini loss caluculation (#5442)

2024-03-11 13:49:58 +08:00 · 2024-03-11 13:49:58 +08:00 · da885ed540
parent 8020f42630
commit da885ed540
1 changed files with 1 additions and 0 deletions
--- a/applications/Colossal-LLaMA-2/train.py
+++ b/applications/Colossal-LLaMA-2/train.py
@ -56,6 +56,7 @@ def format_numel_str(numel: int) -> str:

 def all_reduce_mean(tensor: torch.Tensor) -> torch.Tensor:
    dist.all_reduce(tensor=tensor, op=dist.ReduceOp.SUM)
+    tensor = tensor.data
    tensor.div_(dist.get_world_size())
    return tensor