Mirror/stable-diffusion-webui
1
0
Fork 0
mirror of https://github.com/AUTOMATIC1111/stable-diffusion-webui.git synced 2024-06-07 21:20:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #10201 from brkirch/mps-nan-fixes

Browse Source 
Fix MPS on PyTorch 2.0.1, Intel Macs
This commit is contained in:
AUTOMATIC1111 2023-05-09 10:28:24 +03:00 committed by GitHub
commit ea05ddfec8
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 10 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
modules/mac_specific.py
View File

@ -54,6 +54,11 @@ if has_mps:
CondFunc('torch.cumsum', cumsum_fix_func, None)
CondFunc('torch.Tensor.cumsum', cumsum_fix_func, None)
CondFunc('torch.narrow', lambda orig_func, *args, **kwargs: orig_func(*args, **kwargs).clone(), None)
if version.parse(torch.__version__) == version.parse("2.0"):
# MPS workaround for https://github.com/pytorch/pytorch/issues/96113
CondFunc('torch.nn.functional.layer_norm', lambda orig_func, x, normalized_shape, weight, bias, eps, **kwargs: orig_func(x.float(), normalized_shape, weight.float() if weight is not None else None, bias.float() if bias is not None else bias, eps).to(x.dtype), lambda *args, **kwargs: len(args) == 6)
CondFunc('torch.nn.functional.layer_norm', lambda orig_func, x, normalized_shape, weight, bias, eps, **kwargs: orig_func(x.float(), normalized_shape, weight.float() if weight is not None else None, bias.float() if bias is not None else bias, eps).to(x.dtype), lambda _, input, *args, **kwargs: len(args) == 4 and input.device.type == 'mps')
# MPS workaround for https://github.com/pytorch/pytorch/issues/92311
if platform.processor() == 'i386':
for funcName in ['torch.argmax', 'torch.Tensor.argmax']:
CondFunc(funcName, lambda _, input, *args, **kwargs: torch.max(input.float() if input.dtype == torch.int64 else input, *args, **kwargs)[1], lambda _, input, *args, **kwargs: input.device.type == 'mps')

3
modules/sd_hijack_optimizations.py
View File

@ -256,6 +256,9 @@ def sub_quad_attention_forward(self, x, context=None, mask=None):
k = k.unflatten(-1, (h, -1)).transpose(1,2).flatten(end_dim=1)
v = v.unflatten(-1, (h, -1)).transpose(1,2).flatten(end_dim=1)
if q.device.type == 'mps':
q, k, v = q.contiguous(), k.contiguous(), v.contiguous()
dtype = q.dtype
if shared.opts.upcast_attn:
q, k = q.float(), k.float()