From 5713bf02857c6f7f39c417d89851ea55c87eab97 Mon Sep 17 00:00:00 2001 From: Christian Medina <37550954+cmedinasoriano@users.noreply.github.com> Date: Tue, 30 Jun 2026 22:47:19 -0400 Subject: [PATCH] fix: add CPU offload back to DeepSpeed ZeRO-3 --- training/configs/ornith-35b-lora.yaml | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/training/configs/ornith-35b-lora.yaml b/training/configs/ornith-35b-lora.yaml index 06e9f13..22b8583 100644 --- a/training/configs/ornith-35b-lora.yaml +++ b/training/configs/ornith-35b-lora.yaml @@ -53,6 +53,12 @@ huggingface_hub: deepspeed_config: zero_optimization: stage: 3 + offload_optimizer: + device: cpu + pin_memory: true + offload_param: + device: cpu + pin_memory: true gradient_clipping: 1.0 train_batch_size: auto train_micro_batch_size_per_gpu: auto