| { | |
| "mel": { | |
| "n_fft": 400, | |
| "hop": 160, | |
| "n_mels": 128, | |
| "sample_rate": 16000, | |
| "frames_padded": 3000, | |
| "max_samples": 480000 | |
| }, | |
| "mel_filters_shape": [ | |
| 201, | |
| 128 | |
| ], | |
| "projector": { | |
| "in": 1024, | |
| "out": 512 | |
| }, | |
| "embedding": { | |
| "vocab": 151936, | |
| "dim": 512, | |
| "dtype": "float16" | |
| }, | |
| "lm": { | |
| "num_layers": 8, | |
| "num_heads": 8, | |
| "head_dim": 64, | |
| "hidden": 512, | |
| "max_total_len": 512 | |
| }, | |
| "tokens": { | |
| "user_token": "<|user|>", | |
| "assistant_token": "<|assistant|>", | |
| "bos_audio_token": "<|begin_of_audio|>", | |
| "audio_token": "<|audio|>", | |
| "eos_audio_token": "<|end_of_audio|>", | |
| "assistant_end_token": "<|im_end|>", | |
| "audio_token_id": 151646, | |
| "pad_token_id": 151643, | |
| "eos_token_ids": [ | |
| 151645 | |
| ], | |
| "asr_block_token_id_from": -1, | |
| "extra_block_token_ids": [ | |
| 151647, | |
| 151650, | |
| 151648, | |
| 151646, | |
| 151649 | |
| ], | |
| "vocab_size": 151652 | |
| }, | |
| "prompt": { | |
| "template": "{user}{bos_audio}{audio}{eos_audio}Please transcribe this audio.{assistant}", | |
| "merge_factor": 4 | |
| }, | |
| "generation": { | |
| "max_new_tokens": 128, | |
| "repetition_penalty": 1.0, | |
| "greedy": true | |
| }, | |
| "prompt_ids": { | |
| "user": 151647, | |
| "bos_audio": 151648, | |
| "audio": 151646, | |
| "eos_audio": 151649, | |
| "text": [ | |
| 5501, | |
| 1356, | |
| 3114, | |
| 419, | |
| 7699, | |
| 13 | |
| ], | |
| "assistant": 151650 | |
| }, | |
| "vocab_table": { | |
| "size": 151652 | |
| } | |
| } |