Releases: SWivid/F5-TTS
Releases · SWivid/F5-TTS
1.1.5
What's Changed
- German Model support by @hvoss-techfak in #1072
- Fix finetune training with spaces in file paths by @ionite34 in #1063
- Fix the duration computation in
triton_trtllm/client_grpc.py
by @Nietism in #1071 - Update: Empirically Pruned Step Sampling by @Jerrister in #1077
New Contributors
- @hvoss-techfak made their first contribution in #1072
- @ionite34 made their first contribution in #1063
- @Nietism made their first contribution in #1071
Full Changelog: 1.1.4...1.1.5
1.1.4
1.1.3
Full Changelog: 1.1.0...1.1.3
- v1.1.3 better gradio inference ui, with cherrypick, cache and load text from file support
- v1.1.2 update data prepare script; fix speech edit; fix tensorrt-llm server code-switch inference
- v1.1.1 add benchmark and trtllm offline code
Thanks to @yuekaizhang @fakerybakery @petermg @TBXark
1.1.0
Full Changelog: 1.0.8...1.1.0
- v1.1.0 support GPU depolyment with Triton and TensorRT-LLM by @yuekaizhang in #944
- v1.0.10 support custom chat model by @huanglizhuo in #932
- v1.0.9 several fixes by @huanglizhuo @ZhikangNiu in #924 #926 #928
1.0.8
Full Changelog: 1.0.3...1.0.8
- v1.0.8 fix a fatal bug with log_samples since 37eb3b5 that may be the root cause of training instability and unknown performance degradation
- v1.0.7 add no_zero_init v1 variant path to SHARED.md; add option in finetune gradio to save non-ema model weight
- v1.0.6 improves docker usage and along with other documentation
- v1.0.5 update finetune_gradio.py for clearer guidance
- v1.0.4 fix finetune_gradio.py vocab extend operation with checkpoint in .safetensors format
1.0.3
Full Changelog: 1.0.0...1.0.3
1.0.0
Full Changelog: 0.6.2...1.0.0
Major updates:
- Add text padding mask for both filler tokens and padding tokens.
- Zero-initialize adaptive layernorm in DiT.
- Apply full Rotary Positional Embedding.
Thanks to everyone who has offered advice and support.