Publishing Sad-tts-v1 (`KGSS/Sad-tts-v1`)

Weights — Full OmniVoice-style snapshot (main model.safetensors + audio_tokenizer/). Upstream reference: k2-fsa/OmniVoice.
- model.safetensors
- config.json, tokenizer.json, tokenizer_config.json, chat_template.jinja, etc.
- audio_tokenizer/ (directory with its model.safetensors and configs)
You can sync without Git LFS clutter locally:
```
huggingface-cli download k2-fsa/OmniVoice --local-dir ./mysnap
cp -r miner.py chute_config.yml vocence_config.yaml README.md ./mysnap/
# then push ./mysnap to your Hub repo
```
Revision — Pin the commit SHA in your deploy script as VOCENCE_REVISION and use the same value in vocence miner commit --model-revision.
chute_config.yml on SN78 — Subnet Chutes integrations require TEE + supported GPU class (see your working rhy-TTS-v1 deploy); this folder’s chute_config.yml matches that pattern.
Safetensors metadata — OmniVoice checkpoints from the official Hub use standard PyTorch safetensors metadata; no pt-sn78v-style patching should be needed (unlike some custom Qwen repacks).

Publishing Sad-tts-v1 (KGSS/Sad-tts-v1)