From 9a36452519abd009c3dce998700cf23b01d9e7cf Mon Sep 17 00:00:00 2001 From: Goekdeniz-Guelmez Date: Wed, 5 Mar 2025 14:42:34 +0100 Subject: [PATCH] updates --- llms/mlx_lm/tuner/grpo_trainer.py | 1 + 1 file changed, 1 insertion(+) diff --git a/llms/mlx_lm/tuner/grpo_trainer.py b/llms/mlx_lm/tuner/grpo_trainer.py index 28546cae..098580b1 100644 --- a/llms/mlx_lm/tuner/grpo_trainer.py +++ b/llms/mlx_lm/tuner/grpo_trainer.py @@ -241,6 +241,7 @@ def generate_grpo( if i >= max_tokens - 1: break + mx.metal.clear_cache() mx.eval(current_tokens) if current_tokens: