Modelpublic

might2901/train_super_01

sourceHugging Facecc-by-nc-sa-4.0updated 4mo agoView on Hugging Face

0likes6downloads

Model Card

trainsuper01

Self-distillation LoRA SFT checkpoint (loop 1) for the Vocence subnet.

—Base: gold24k/vocence-new-tts
—Training data: eval_train.jsonl (validator-style prompts + miner audio)
—Checkpoint: epoch 1 merged weights

Prompt-driven English TTS on Qwen3-TTS-12Hz-1.7B-VoiceDesign. Served via canonical miner.py on Chutes.

python

from qwen_tts import Qwen3TTSModel
m = Qwen3TTSModel.from_pretrained("might2901/train_super_01")
wavs, sr = m.generate_voice_design(
    text="Hello world.",
    instruct="A calm adult male with an American accent.",
    language="english",
)

CC BY-NC-SA 4.0.