params.json
| 1 | { |
| 2 | "dim": 3072, |
| 3 | "n_layers": 26, |
| 4 | "head_dim": 128, |
| 5 | "hidden_dim": 9216, |
| 6 | "n_heads": 32, |
| 7 | "n_kv_heads": 8, |
| 8 | "use_biases": false, |
| 9 | "causal": true, |
| 10 | "rope_theta": 1000000.0, |
| 11 | "norm_eps": 1e-05, |
| 12 | "vocab_size": 131072, |
| 13 | "model_parallel": 1, |
| 14 | "tied_embeddings": true, |
| 15 | "sliding_window": 8192, |
| 16 | "model_max_length": 131072, |
| 17 | "multimodal": { |
| 18 | "whisper_model_args": { |
| 19 | "encoder_args": { |
| 20 | "audio_encoding_args": { |
| 21 | "sampling_rate": 16000, |
| 22 | "frame_rate": 12.5, |
| 23 | "num_mel_bins": 128, |
| 24 | "hop_length": 160, |
| 25 | "window_size": 400, |
| 26 | "chunk_length_s": null, |
| 27 | "global_log_mel_max": 1.5, |
| 28 | "transcription_format": "streaming" |
| 29 | }, |
| 30 | "dim": 1280, |
| 31 | "n_layers": 32, |
| 32 | "head_dim": 64, |
| 33 | "hidden_dim": 5120, |
| 34 | "n_heads": 32, |
| 35 | "vocab_size": 131072, |
| 36 | "n_kv_heads": 32, |
| 37 | "use_biases": true, |
| 38 | "use_cache": false, |
| 39 | "rope_theta": 1000000.0, |
| 40 | "causal": true, |
| 41 | "norm_eps": 1e-05, |
| 42 | "pos_embed": "rope", |
| 43 | "max_source_positions": null, |
| 44 | "ffn_type": "swiglu", |
| 45 | "norm_type": "rms_norm", |
| 46 | "sliding_window": 750 |
| 47 | }, |
| 48 | "downsample_args": { |
| 49 | "downsample_factor": 4 |
| 50 | } |
| 51 | } |
| 52 | }, |
| 53 | "ada_rms_norm_t_cond": true, |
| 54 | "ada_rms_norm_t_cond_dim": 32 |
| 55 | } |
| 56 | |