mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-07-04 02:57:20 +00:00
[bug] fixed broken test_found_inf (#725)
This commit is contained in:
parent
193dc8dacb
commit
20ab1f5520
@ -31,7 +31,7 @@ def _run_test_found_inf(cpu_offload, shard_strategy_class, gpu_margin_mem_ratio)
|
|||||||
model_builder, train_dataloader, _, optimizer_class, criterion = get_components_func()
|
model_builder, train_dataloader, _, optimizer_class, criterion = get_components_func()
|
||||||
|
|
||||||
with ZeroInitContext(
|
with ZeroInitContext(
|
||||||
target_device=torch.device(f'cpu:0') if cpu_offload else torch.device(f'cuda:{get_current_device()}'),
|
target_device=torch.device(f'cpu:0') if cpu_offload else torch.device(get_current_device()),
|
||||||
shard_strategy=shard_strategy,
|
shard_strategy=shard_strategy,
|
||||||
shard_param=True):
|
shard_param=True):
|
||||||
zero_model = model_builder(checkpoint=True)
|
zero_model = model_builder(checkpoint=True)
|
||||||
|
Loading…
Reference in New Issue
Block a user