Skip to content

Fix for Mask Patch Failure and Quantization Issues in Latest transformers Versions #368

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

Merged
merged 4 commits into from
Jul 3, 2025
Merged

Fix for Mask Patch Failure and Quantization Issues in Latest transformers Versions #368

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2447030
Update torch2coreml.py
Aintor Nov 10, 2024
4ef83f1
Update torch2coreml.py
Aintor Nov 12, 2024
7e3df29
Update torch2coreml.py
Aintor Nov 12, 2024
50bb4c4
Update torch2coreml.py
Aintor Nov 12, 2024
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
7 changes: 5 additions & 2 deletions python_coreml_stable_diffusion/torch2coreml.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,6 +322,8 @@ def bundle_resources_for_swift_cli(args):
from transformers.models.clip import modeling_clip

# Copied from https://github.com/huggingface/transformers/blob/v4.30.0/src/transformers/models/clip/modeling_clip.py#L677C1-L692C1
# Starting from transformers >= 4.35.0, the _make_causal_mask function is replaced by _create_4d_causal_attention_mask in modeling_clip.
# For backward compatibility with versions < 4.35.0, both functions are patched here.
def patched_make_causal_mask(input_ids_shape, dtype, device, past_key_values_length: int = 0):
""" Patch to replace torch.finfo(dtype).min with -1e4
"""
Expand All @@ -334,8 +336,9 @@ def patched_make_causal_mask(input_ids_shape, dtype, device, past_key_values_len
if past_key_values_length > 0:
mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
return mask[None, None, :, :].expand(bsz, 1, tgt_len, tgt_len + past_key_values_length)

modeling_clip._make_causal_mask = patched_make_causal_mask

modeling_clip._make_causal_mask = patched_make_causal_mask # For transformers >= 4.30.0 and transformers < 4.35.0
modeling_clip._create_4d_causal_attention_mask = patched_make_causal_mask # For transformers >= 4.35.0

def convert_text_encoder(text_encoder, tokenizer, submodule_name, args):
""" Converts the text encoder component of Stable Diffusion
Expand Down