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

Add a weight_dtype fp8_e4m3fn_fast to the Diffusion Model Loader node.

Browse Source 
This is used to load weights in fp8 and use fp8 matrix multiplication.
This commit is contained in:
comfyanonymous
2024-10-09 19:43:17 -04:00
parent 203942c8b2
commit e38c94228b
6 changed files with 27 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
comfy/ops.py
View File

@@ -299,7 +299,11 @@ class fp8_ops(manual_cast):
return torch.nn.functional.linear(input, weight, bias)
def pick_operations(weight_dtype, compute_dtype, load_device=None, disable_fast_fp8=False):
def pick_operations(weight_dtype, compute_dtype, load_device=None, disable_fast_fp8=False, fp8_optimizations=False):
if comfy.model_management.supports_fp8_compute(load_device):
if (fp8_optimizations or args.fast) and not disable_fast_fp8:
return fp8_ops
if compute_dtype is None or weight_dtype == compute_dtype:
return disable_weight_init
if args.fast and not disable_fast_fp8: