From 0456d55c8adf5ed72b589330e2d9ecf73561fb2b Mon Sep 17 00:00:00 2001 From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com> Date: Thu, 2 Jul 2026 17:29:19 -0400 Subject: [PATCH] fix: disable gradient checkpointing (device mismatch with distributed MoE) --- training/configs/ornith-35b-lora.yaml | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/training/configs/ornith-35b-lora.yaml b/training/configs/ornith-35b-lora.yaml index f9ac3a1..b43acf5 100644 --- a/training/configs/ornith-35b-lora.yaml +++ b/training/configs/ornith-35b-lora.yaml @@ -53,5 +53,5 @@ mixed_precision: bf16 # Evaluation (disable - no test split in dataset) eval_strategy: "no" -# Gradient Checkpointing -gradient_checkpointing: true +# Gradient Checkpointing (disable - causes device issues with distributed MoE) +gradient_checkpointing: false