jinx/ComfyUI
1
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-08-02 23:14:49 +08:00
Code Issues Activity

Allow FP16 accumulation with --fast (#6453)

Browse Source 
Currently only applies to PyTorch nightly releases. (>=20250208)
This commit is contained in:
catboxanon
2025-02-08 17:00:56 -05:00
committed by GitHub
parent af93c8d1ee
commit 43a74c0de1
1 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
comfy/model_management.py
View File

@@ -241,6 +241,12 @@ if ENABLE_PYTORCH_ATTENTION:
torch.backends.cuda.enable_flash_sdp(True)
torch.backends.cuda.enable_mem_efficient_sdp(True)
try:
if is_nvidia() and args.fast:
torch.backends.cuda.matmul.allow_fp16_accumulation = True
except:
pass
try:
if int(torch_version[0]) == 2 and int(torch_version[2]) >= 5:
torch.backends.cuda.allow_fp16_bf16_reduction_math_sdp(True)