From 7691cbc2dc0d72f5da6411b0fe65bc29a3a5515f Mon Sep 17 00:00:00 2001 From: lucasdelimanogueira Date: Mon, 3 Jun 2024 15:15:35 -0300 Subject: [PATCH] fix allreduce distributed not working after 2nd iteration --- norch/nn/parallel.py | 1 - 1 file changed, 1 deletion(-) diff --git a/norch/nn/parallel.py b/norch/nn/parallel.py index 8d6fef9..338ef18 100644 --- a/norch/nn/parallel.py +++ b/norch/nn/parallel.py @@ -32,7 +32,6 @@ class DistributedDataParallel(Module): if isinstance(grad, norch.Tensor): dist.allreduce_sum_tensor(grad) avg_grad = grad / dist.get_world_size() - avg_grad = 0*grad + 5 return avg_grad def register_grads_hooks(self):