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

fix: remove sync_module_states=True (model is on CPU)

Browse Source
This commit is contained in:
Christian Medina
2026-07-02 11:55:21 -04:00
parent 5a13ca1d1c
commit 14ef1a07c4
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
train.py
View File

@@ -201,7 +201,7 @@ def train(config_path):
auto_wrap_policy=get_auto_wrap_policy(model), auto_wrap_policy=get_auto_wrap_policy(model),
device_id=None, # Keep on CPU initially device_id=None, # Keep on CPU initially
mixed_precision=None, mixed_precision=None,
sync_module_states=True, sync_module_states=False, # Model is on CPU, no sync needed
use_orig_params=True, use_orig_params=True,
) )
print("✓ Model wrapped with FSDP (will be sharded across GPUs during training)") print("✓ Model wrapped with FSDP (will be sharded across GPUs during training)")