Skip to content

fix: update XTTS/Tortoise GPT code for HF transformers 4.52+ #414

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
Merged
merged 8 commits into from
Jun 12, 2025
Merged

fix: update XTTS/Tortoise GPT code for HF transformers 4.52+ #414

merged 8 commits into from
Jun 12, 2025

Conversation

eginhard
Copy link
Member

@eginhard eginhard commented Jun 3, 2025 •
edited
Loading

  • adapt to HF changes that moved a lot of code from generate() into separate helper methods, so we don't need to duplicate that code in stream_generator.py anymore and can just call these methods
  • remove some obsolete code to fall back to official HF implementations that have e.g. improved caching support
  • remove branches like beam-search in xtts/stream_generator.py that are never used to shorten that code
  • minimum transformers version increased to 4.52

Fixes #400

@eginhard eginhard requested review from KarlHajal and removed request for KarlHajal June 3, 2025 19:07
@eginhard eginhard marked this pull request as draft June 3, 2025 22:58
@eginhard eginhard mentioned this pull request Jun 4, 2025
Closed
@eginhard eginhard changed the title fix(xtts): remove now obsolete code for transformer 4.52 compatibility fix: update XTTS/Tortoise GPT code for HF transformers 4.50+ Jun 4, 2025
@eginhard eginhard requested a review from KarlHajal June 5, 2025 08:10
@eginhard eginhard marked this pull request as ready for review June 5, 2025 08:10
@eginhard eginhard changed the title fix: update XTTS/Tortoise GPT code for HF transformers 4.50+ fix: update XTTS/Tortoise GPT code for HF transformers 4.52+ Jun 5, 2025
eginhard added 8 commits June 12, 2025 01:07
@eginhard
fix(xtts): pass attention mask also for streaming tts
dc185a9
@eginhard
fix(xtts): adapt gpt code to transformers 4.50+
8555c60 
- adapt to HF changes that moved a lot of code from generate() into separate
helper methods, so we don't need to duplicate that code in stream_generator.py
anymore
- removes some obsolete code to fall back to official HF implementations that
have e.g. improved caching support
- remove branches like beam-search in stream_generator.py that are never used to
shorten that code
@eginhard
fix(tortoise): adapt gpt code to transformers 4.50+
2bdc21e
@eginhard
refactor(bark): remove version checks around flash attention
93f9d88 
scaled_dot_product_attention is now available in all Pytorch versions we support
@eginhard
fix(bark): add more detailed logs for model downloads
7affa3d
@eginhard
docs(faq): add info about the fork and fine-tuning
27f87f7
@eginhard
Revert "build: exclude fugashi 1.5.0"
a72a09a 
This reverts commit 75cc42a.

Obsolete because 1.5.0 has been yanked
@eginhard
fix(bark): disable weights-only loading
7485e4e 
It doesn't seem possible to reliably do this with Numpy>2 because of
np._core.multiarray.scalar
@eginhard eginhard merged commit d59aca3 into dev Jun 12, 2025
30 checks passed
@eginhard eginhard deleted the tf452 branch June 12, 2025 07:48
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@KarlHajal KarlHajal KarlHajal approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

XTTS compatibility with transformers>=4.52
2 participants
@eginhard @KarlHajal