| { | |
| "mlx_format_version": 1, | |
| "encoder": { | |
| "encoder_dim": 1280, | |
| "n_mels": 128, | |
| "num_layers": 32, | |
| "num_heads": 20, | |
| "head_dim": 64, | |
| "intermediate_size": 5120, | |
| "rope_theta": 10000.0 | |
| }, | |
| "projector": { | |
| "encoder_dim": 1280, | |
| "llm_dim": 1024, | |
| "hidden_dim": 1024, | |
| "pool_stride": 4 | |
| }, | |
| "audio_token": "<audio>", | |
| "audio_token_id": 151669, | |
| "eos_token_ids": [ | |
| 151643, | |
| 151645 | |
| ], | |
| "encoder_conv_layers": [ | |
| [ | |
| 1, | |
| 3, | |
| 1 | |
| ], | |
| [ | |
| 1, | |
| 3, | |
| 2 | |
| ] | |
| ], | |
| "hop_length": 160 | |
| } |