| { | |
| "attention_backend": "auto", | |
| "attention_window": [ | |
| 64, | |
| 64 | |
| ], | |
| "codebook_dim": 8, | |
| "codebook_size": 16384, | |
| "commitment": 0.5, | |
| "decoder_layers_stage1": 6, | |
| "decoder_layers_stage2": 12, | |
| "dim": 512, | |
| "dropout": 0.0, | |
| "encoder_layers_stage1": 6, | |
| "encoder_layers_stage2": 12, | |
| "ffn_mult": 4, | |
| "hop_length": 80, | |
| "initial_tau": 0.19603846967220306, | |
| "keep_ratio": 0.5, | |
| "layerscale_gamma_init": 1.0, | |
| "max_position_embeddings": 8192, | |
| "n_fft": 320, | |
| "norm_eps": 0.01, | |
| "num_heads": 4, | |
| "num_quantizers": 1, | |
| "rope_base": 10000.0, | |
| "sample_rate": 16000, | |
| "window_size": 320 | |
| } | |