reazon-research/japanese-wav2vec2-base-rs35kh

Full fine-tune vs reazon-research/japanese-wav2vec2-base. Fine-tune: 96% of the weights changed, across 12 of 12 layers.

Tensors changed: Norms 52/54, Attention 96/96, MLP 48/48, Embeddings 3/3, other 2/9

On the Hub · Downloads on Model Pulse