Let all model memory be offloaded on nvidia.
This commit is contained in:
@@ -220,7 +220,7 @@ def is_amd():
|
|||||||
|
|
||||||
MIN_WEIGHT_MEMORY_RATIO = 0.4
|
MIN_WEIGHT_MEMORY_RATIO = 0.4
|
||||||
if is_nvidia():
|
if is_nvidia():
|
||||||
MIN_WEIGHT_MEMORY_RATIO = 0.1
|
MIN_WEIGHT_MEMORY_RATIO = 0.0
|
||||||
|
|
||||||
ENABLE_PYTORCH_ATTENTION = False
|
ENABLE_PYTORCH_ATTENTION = False
|
||||||
if args.use_pytorch_cross_attention:
|
if args.use_pytorch_cross_attention:
|
||||||
|
|||||||
Reference in New Issue
Block a user