ros/nanochat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: remove unnecessary tensor allocation in DistAdamW optimizer

Browse Source 
fix: remove unnecessary tensor allocation in DistAdamW optimizer
This commit is contained in:
Sermet Pekin
2025-10-20 12:03:26 +03:00
committed by GitHub
parent 0f007889dd
commit 49cd02f283
1 changed files with 0 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
nanochat/adamw.py
View File

@@ -26,7 +26,6 @@ class DistAdamW(torch.optim.Optimizer):
grad_slices = [] grad_slices = []
for group in self.param_groups: for group in self.param_groups:
params: list[Tensor] = group["params"] params: list[Tensor] = group["params"]
grad = torch.empty_like(params[-1]) # TODO is this bug? seems to be over-written instantly
for base_i in range(len(params)): for base_i in range(len(params)):
grad = params[base_i].grad grad = params[base_i].grad
rank_size = grad.shape[0] // world_size rank_size = grad.shape[0] // world_size