RAM optimizations and faster launch
This commit is contained in:
@@ -496,7 +496,7 @@ class T5EncoderModel:
|
||||
device=device).eval().requires_grad_(False)
|
||||
logging.info(f'loading {checkpoint_path}')
|
||||
from mmgp import offload
|
||||
offload.load_model_data(model,checkpoint_path )
|
||||
offload.load_model_data(model,checkpoint_path, writable_tensors= False )
|
||||
|
||||
self.model = model
|
||||
if shard_fn is not None:
|
||||
|
||||
Reference in New Issue
Block a user