Vocence TTS miner snapshot.
model.safetensors header re-stamped with model_id=God-tts-v1 / build_tag=god-v1-2026-05-11
so it diverges from any sibling snapshot's header hash, while the tensor payload
remains bit-identical to the base Qwen3-TTS-12Hz-1.7B-CustomVoice fine-tune.
training_state.pt (optimizer state, 11.5 GB) intentionally omitted; chute inference does not need it.