Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Add FP32 fallback support on sd_vae_approx #14046

Merged
merged 4 commits into from
Dec 2, 2023
Merged

Add FP32 fallback support on sd_vae_approx #14046

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
58c1954
Add FP32 fallback support on sd_vae_approx
hidenorly Nov 20, 2023
39eae9f
Revert "Add FP32 fallback support on sd_vae_approx"
hidenorly Nov 28, 2023
a0096c5
Add FP32 fallback support on torch.nn.functional.interpolate
hidenorly Nov 28, 2023
81c0072
Fix the Ruff error about unused import
hidenorly Nov 28, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions modules/mac_specific.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
import logging

import torch
from torch import Tensor
import platform
from modules.sd_hijack_utils import CondFunc
from packaging import version
Expand Down Expand Up @@ -51,6 +52,17 @@ def cumsum_fix(input, cumsum_func, *args, **kwargs):
return cumsum_func(input, *args, **kwargs)


# MPS workaround for https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046
def interpolate_with_fp32_fallback(orig_func, *args, **kwargs) -> Tensor:
try:
return orig_func(*args, **kwargs)
except RuntimeError as e:
if "not implemented for" in str(e) and "Half" in str(e):
input_tensor = args[0]
return orig_func(input_tensor.to(torch.float32), *args[1:], **kwargs).to(input_tensor.dtype)
else:
print(f"An unexpected RuntimeError occurred: {str(e)}")

if has_mps:
if platform.mac_ver()[0].startswith("13.2."):
# MPS workaround for https://github.com/pytorch/pytorch/issues/95188, thanks to danieldk (https://github.com/explosion/curated-transformers/pull/124)
Expand All @@ -77,6 +89,9 @@ def cumsum_fix(input, cumsum_func, *args, **kwargs):
# 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 _, input, *args, **kwargs: len(args) == 4 and input.device.type == 'mps')

# MPS workaround for https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/14046
CondFunc('torch.nn.functional.interpolate', interpolate_with_fp32_fallback, None)

# MPS workaround for https://github.com/pytorch/pytorch/issues/92311
if platform.processor() == 'i386':
for funcName in ['torch.argmax', 'torch.Tensor.argmax']:
Expand Down