v3.1.0

What's Changed

• Document CPU-only installation (pip pulls ~3 GB of CUDA wheels by default) by @jflaflamme in #219
• Speed table: measured H100 rows off local storage by @manu-lm in #250
• lsd_depth_distill: 64 rows in one pass, drop the unused flow multiplier by @manu-lm in #253
• Finetune a released model into a new language by @manu-lm in #254
• Import Annotated from typing by @manu-lm in #256
• Run the training tests in CI, and make zip lengths explicit by @manu-lm in #257
• Run the training tests in CI by @manu-lm in #258
• Fix decimal splitting in sentence chunker by @preciousimo in #217
• Split tts_model into text chunking and model-state I/O by @manu-lm in #259
• Immutable defaults in seanet; document the two open() call sites by @manu-lm in #260
• CONTRIBUTING: encourage sharing trained models by @manu-lm in #266
• Add a --default-voice option to the serve command by @gabrieldemarmiesse in #271
• README: explain depth+CFG distillation by @manu-lm in #272
• Move text conditioning into modules/text_conditioner.py by @manu-lm in #275
• Use tanh GELU approximation in the backbone FFN by @manu-lm in #278
• Replace beartype with ty for type checking by @gabrieldemarmiesse-ai-agent in #283
• Add Hindi community model and configurable terminal punctuation by @Sedherthe in #288
• Add "pocket-tts-timestamped" project to README by @dpm63 in #290
• Add Korean 300M community model to README by @seastar105 in #291
• README: Czech example with a Czech sentence by @manu-lm in #292

New Contributors

• @jflaflamme made their first contribution in #219
• @preciousimo made their first contribution in #217
• @gabrieldemarmiesse-ai-agent made their first contribution in #263
• @Sedherthe made their first contribution in #288
• @dpm63 made their first contribution in #290
• @seastar105 made their first contribution in #291

Full Changelog: v3.0.2...v3.1.0