cmedina/agenx-lora-training
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: add CPU offload back to DeepSpeed ZeRO-3

Browse Source
This commit is contained in:
Christian Medina
2026-06-30 22:47:19 -04:00
parent 3bb7328e2a
commit 5713bf0285
1 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
training/configs/ornith-35b-lora.yaml
View File

@@ -53,6 +53,12 @@ huggingface_hub:
deepspeed_config: deepspeed_config:
zero_optimization: zero_optimization:
stage: 3 stage: 3
offload_optimizer:
device: cpu
pin_memory: true
offload_param:
device: cpu
pin_memory: true
gradient_clipping: 1.0 gradient_clipping: 1.0
train_batch_size: auto train_batch_size: auto
train_micro_batch_size_per_gpu: auto train_micro_batch_size_per_gpu: auto