kjanh/KhanhTTS-OmniVoice

Full fine-tune vs k2-fsa/OmniVoice. Fine-tune: 100% of the weights changed, across 28 of 28 layers.

Tensors changed: other 1/2, Embeddings 2/2, Norms 113/113, MLP 84/84, Attention 112/112

5 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse