wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-02-12 22:42:36 +08:00
Code Issues Actions 20 Packages Projects Releases Wiki Activity

Merge branch 'comfyanonymous:master' into fix/secure-combo

Browse Source
This commit is contained in:
Dr.Lt.Data 2023-07-03 13:33:55 +09:00 committed by GitHub
commit 4a8d87f175
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 11 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
comfy/ldm/modules/attention.py
View File

@ -16,11 +16,14 @@ if model_management.xformers_enabled():
import xformers import xformers
import xformers.ops import xformers.ops
# CrossAttn precision handling
import os
_ATTN_PRECISION = os.environ.get("ATTN_PRECISION", "fp32")
from comfy.cli_args import args from comfy.cli_args import args
# CrossAttn precision handling
if args.dont_upcast_attention:
print("disabling upcasting of attention")
_ATTN_PRECISION = "fp16"
else:
_ATTN_PRECISION = "fp32"
def exists(val): def exists(val):
return val is not None return val is not None

8
comfy/model_management.py
View File

@ -334,19 +334,19 @@ def unload_if_low_vram(model):
return model return model
def unet_offload_device(): def unet_offload_device():
if vram_state == VRAMState.HIGH_VRAM or vram_state == VRAMState.SHARED: if vram_state == VRAMState.HIGH_VRAM:
return get_torch_device() return get_torch_device()
else: else:
return torch.device("cpu") return torch.device("cpu")
def text_encoder_offload_device(): def text_encoder_offload_device():
if args.gpu_only or vram_state == VRAMState.SHARED: if args.gpu_only:
return get_torch_device() return get_torch_device()
else: else:
return torch.device("cpu") return torch.device("cpu")
def text_encoder_device(): def text_encoder_device():
if args.gpu_only or vram_state == VRAMState.SHARED: if args.gpu_only:
return get_torch_device() return get_torch_device()
elif vram_state == VRAMState.HIGH_VRAM or vram_state == VRAMState.NORMAL_VRAM: elif vram_state == VRAMState.HIGH_VRAM or vram_state == VRAMState.NORMAL_VRAM:
if torch.get_num_threads() < 8: #leaving the text encoder on the CPU is faster than shifting it if the CPU is fast enough. if torch.get_num_threads() < 8: #leaving the text encoder on the CPU is faster than shifting it if the CPU is fast enough.
@ -360,7 +360,7 @@ def vae_device():
return get_torch_device() return get_torch_device()
def vae_offload_device(): def vae_offload_device():
if args.gpu_only or vram_state == VRAMState.SHARED: if args.gpu_only:
return get_torch_device() return get_torch_device()
else: else:
return torch.device("cpu") return torch.device("cpu")

4
main.py
View File

@ -14,10 +14,6 @@ if os.name == "nt":
logging.getLogger("xformers").addFilter(lambda record: 'A matching Triton is not available' not in record.getMessage()) logging.getLogger("xformers").addFilter(lambda record: 'A matching Triton is not available' not in record.getMessage())
if __name__ == "__main__": if __name__ == "__main__":
if args.dont_upcast_attention:
print("disabling upcasting of attention")
os.environ['ATTN_PRECISION'] = "fp16"
if args.cuda_device is not None: if args.cuda_device is not None:
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.cuda_device) os.environ['CUDA_VISIBLE_DEVICES'] = str(args.cuda_device)
print("Set cuda device to:", args.cuda_device) print("Set cuda device to:", args.cuda_device)