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