From 3ee548cd06445d559e23e72942dea9675df5058a Mon Sep 17 00:00:00 2001 From: Benoit <69694610+BenoitDalFerro@users.noreply.github.com> Date: Thu, 19 Sep 2024 19:18:55 +0200 Subject: [PATCH] Update AdEMAMix-Shampoo.py Fixing division by 0 error at first step --- AdEMAMix-Shampoo.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/AdEMAMix-Shampoo.py b/AdEMAMix-Shampoo.py index 2a73931..d2e5dfb 100644 --- a/AdEMAMix-Shampoo.py +++ b/AdEMAMix-Shampoo.py @@ -284,7 +284,7 @@ def _update_adamemix_distributed_shampoo( denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(eps) # Compute step size - step_size = lr / bias_correction1 + step_size = lr / (bias_correction1 if bias_correction1 > 0 else 0.01) # Apply weight decay if weight_decay != 0: