From 601c6d6aa8781e5198e80852c8ac8077ba9c1f70 Mon Sep 17 00:00:00 2001 From: David Koski <46639364+davidkoski@users.noreply.github.com> Date: Thu, 1 Feb 2024 09:56:27 -0800 Subject: [PATCH] Fix for AdaDelta (#603) - state was being read from parameter "s" - but being stored in parameter "u" --- python/mlx/optimizers.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/mlx/optimizers.py b/python/mlx/optimizers.py index c47234bd5..b659ec5cf 100644 --- a/python/mlx/optimizers.py +++ b/python/mlx/optimizers.py @@ -284,7 +284,7 @@ class AdaDelta(Optimizer): eps = self.eps v = state.get("v", mx.zeros_like(gradient)) - u = state.get("s", mx.zeros_like(gradient)) + u = state.get("u", mx.zeros_like(gradient)) v = rho * v + (1 - rho) * mx.square(gradient) d = mx.sqrt(u + eps) / mx.sqrt(v + eps) * gradient