298977887/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-01-11 02:15:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Replace use of .view with .reshape (#4522)

Browse Source 
When generating images with fp8_e4_m3 Flux and batch size >1, using --fast, ComfyUI throws a "view size is not compatible with input tensor's size and stride" error pointing at the first of these two calls to view.

As reshape is semantically equivalent to view except for working on a broader set of inputs, there should be no downside to changing this. The only difference is that it clones the underlying data in cases where .view would error out. I have confirmed that the output still looks as expected, but cannot confirm that no mutable use is made of the tensors anywhere.

Note that --fast is only marginally faster than the default.
This commit is contained in:
Svein Ove Aas 2024-08-21 16:21:48 +01:00 committed by GitHub
parent 5e806f555d
commit 5f50263088
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
comfy/ops.py
View File

@ -250,7 +250,7 @@ def fp8_linear(self, input):
return None
if len(input.shape) == 3:
inn = input.view(-1, input.shape[2]).to(dtype)
inn = input.reshape(-1, input.shape[2]).to(dtype)
non_blocking = comfy.model_management.device_supports_non_blocking(input.device)
w = cast_to(self.weight, device=input.device, non_blocking=non_blocking).t()
@ -259,7 +259,7 @@ def fp8_linear(self, input):
else:
o, _ = torch._scaled_mm(inn, w, out_dtype=input.dtype)
return o.view((-1, input.shape[1], self.weight.shape[0]))
return o.reshape((-1, input.shape[1], self.weight.shape[0]))
return None
class fp8_ops(manual_cast):