wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-12-16 17:42:58 +08:00
Code Issues Actions 13 Packages Projects Releases Wiki Activity

Reduce memory usage for fp8 scaled op. (#10531)

Browse Source
This commit is contained in:
comfyanonymous 2025-10-29 12:43:51 -07:00 committed by GitHub
parent 6c14f3afac
commit 1a58087ac2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
comfy/quant_ops.py
View File

@ -358,7 +358,7 @@ class TensorCoreFP8Layout(QuantizedLayout):
scale = scale.to(device=tensor.device, dtype=torch.float32) scale = scale.to(device=tensor.device, dtype=torch.float32)
lp_amax = torch.finfo(dtype).max lp_amax = torch.finfo(dtype).max
tensor_scaled = tensor.float() / scale tensor_scaled = tensor * (1.0 / scale).to(tensor.dtype)
torch.clamp(tensor_scaled, min=-lp_amax, max=lp_amax, out=tensor_scaled) torch.clamp(tensor_scaled, min=-lp_amax, max=lp_amax, out=tensor_scaled)
qdata = tensor_scaled.to(dtype, memory_format=torch.contiguous_format) qdata = tensor_scaled.to(dtype, memory_format=torch.contiguous_format)