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

Support musicgen conversion. #296

Merged
merged 1 commit into from
Jul 11, 2023
+18 −3
Merged

Support musicgen conversion. #296

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
21 changes: 18 additions & 3 deletions bindings/python/convert.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,8 +161,11 @@ def check_final_model(model_id: str, folder: str):
shutil.copy(config, os.path.join(folder, "config.json"))
config = AutoConfig.from_pretrained(folder)

_, (pt_model, pt_infos) = infer_framework_load_model(model_id, config, output_loading_info=True)
_, (sf_model, sf_infos) = infer_framework_load_model(folder, config, output_loading_info=True)
import transformers

class_ = getattr(transformers, config.architectures[0])
(pt_model, pt_infos) = class_.from_pretrained(folder, output_loading_info=True)
(sf_model, sf_infos) = class_.from_pretrained(folder, output_loading_info=True)
Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The old version doesn't work anymore.

This might fail on very old models (where config.architectures is not written), but I think it's ok now, they should be converted by now.


if pt_infos != sf_infos:
error_string = create_diff(pt_infos, sf_infos)
Expand Down Expand Up @@ -199,7 +202,19 @@ def check_final_model(model_id: str, folder: str):
sf_model = sf_model.cuda()
kwargs = {k: v.cuda() for k, v in kwargs.items()}

pt_logits = pt_model(**kwargs)[0]
try:
pt_logits = pt_model(**kwargs)[0]
except Exception as e:
try:
# Musicgen special exception.
decoder_input_ids = torch.ones((input_ids.shape[0] * pt_model.decoder.num_codebooks, 1), dtype=torch.long)
if torch.cuda.is_available():
decoder_input_ids = decoder_input_ids.cuda()

kwargs["decoder_input_ids"] = decoder_input_ids
pt_logits = pt_model(**kwargs)[0]
except Exception:
raise e
sf_logits = sf_model(**kwargs)[0]

torch.testing.assert_close(sf_logits, pt_logits)
Expand Down