| accum_grad: 1 | |
| cmvn_file: | |
| dataset_conf: | |
| batch_conf: | |
| batch_size: 4 | |
| batch_type: chat | |
| fbank_conf: | |
| chunk_size: 4 | |
| dither: 0.0 | |
| frame_length: 25 | |
| frame_shift: 10 | |
| num_mel_bins: 80 | |
| pad_rate: 0.1 | |
| filter_conf: | |
| max_length: 2000 | |
| max_output_input_ratio: 102400 | |
| max_total_num: 1000 | |
| min_length: 10 | |
| min_output_input_ratio: 0 | |
| token_max_length: 102400 | |
| token_min_length: 0 | |
| parse_multi_rounds: true | |
| resample_conf: | |
| resample_rate: 16000 | |
| shuf_list: true | |
| shuffle: false | |
| sort: false | |
| spec_aug: true | |
| spec_aug_conf: | |
| max_f: 10 | |
| max_t: 20 | |
| num_f_mask: 1 | |
| num_t_mask: 1 | |
| spec_sub: false | |
| spec_sub_conf: | |
| max_t: 30 | |
| num_t_sub: 3 | |
| speed_perturb: false | |
| split_num: 1 | |
| tokenize_char: false | |
| tokenize_conf: | |
| eod_id: 151645 | |
| tokenize_type: Qwen | |
| tokenizer_path: | |
| ds_dtype: bf16 | |
| encoder_conf: | |
| overview_conf: | |
| encoder-input-dim: 80 | |
| encoder-layer-config: subsampling-transformer | |
| encoder-output-dim: 1024 | |
| para_conf: | |
| subsampling: | |
| subsampling-dropout-rate: 0.1 | |
| subsampling-input-dim: 80 | |
| subsampling-output-dim: 1024 | |
| subsampling-rate: 4 | |
| transformer: | |
| transformer-attention-dim: 1024 | |
| transformer-attention-dropout-rate: 0.0 | |
| transformer-attention-heads: 16 | |
| transformer-chunk_size: 4 | |
| transformer-concat-after: false | |
| transformer-dropout-rate: 0.1 | |
| transformer-dynamic-chunks: false | |
| transformer-input-dim: 1024 | |
| transformer-input-layer: linear | |
| transformer-left_chunks: 16 | |
| transformer-linear-units: 4096 | |
| transformer-normalize-before: true | |
| transformer-num-blocks: 24 | |
| transformer-output-dim: 1024 | |
| transformer-pos-enc-class: rel-enc | |
| transformer-positional-dropout-rate: 0.1 | |
| transformer-positionwise-layer-type: linear | |
| grad_clip: 5.0 | |
| input_dim: 80 | |
| is_json_cmvn: true | |
| lang_dict: | |
| lfmmi_dir: '' | |
| log_interval: 100 | |
| max_epoch: 100 | |
| model_conf: | |
| activation_func: gelu | |
| add_audio_bos_eos: true | |
| add_prompt_before: true | |
| adpter_type: subsampling | |
| chat_template: '<|im_start|>system | |
| You are a helpful assistant.<|im_end|> | |
| <|im_start|>user | |
| <audio><|im_end|> | |
| <|im_start|>assistant | |
| ' | |
| chunk_size: 2 | |
| enc_out_dim: 1024 | |
| freeze_adpter: true | |
| freeze_encoder: true | |
| freeze_llm: true | |
| kernel_size: 5 | |
| llm_embed_dim: 3584 | |
| llm_head_num: 28 | |
| llm_path: | |
| norm: layer | |
| num_key_value_heads: 4 | |
| predict_usr_state: 4 | |
| prompt_finetune: true | |
| prompt_num: 25 | |
| task_num: 20 | |
| task_type: prompt_finetune | |
| optim: adamw | |
| optim_conf: | |
| betas: | |
| - 0.9 | |
| - 0.99 | |
| eps: 1.0e-06 | |
| lr: 0.0006 | |
| weight_decay: 0.01 | |
| output_dim: 5538 | |
| scheduler: warmuplr | |
| scheduler_conf: | |
| warmup_steps: 200 | |