MOSS-TTS-Nano Pruned
A pruned version of MOSS-TTS-Nano-100M, optimized for lower memory usage and faster inference.
Base Model
OpenMOSS-Team/MOSS-TTS-Nano-100M.
Key Changes
- Global Transformer layers reduced to 6
- Local Transformer layers reduced by approximately 50%
- Audio VQ codebooks reduced to 8
Note
Fine-tuning is required to recover the original model's multilingual capabilities and zero-shot voice cloning performance.
- Downloads last month
- -