[
    896,
    1024,
    {
        "accum_grad": 3,
        "char_list": [],
        "debugmode": 0,
        "encoder_criterion": "ce",
        "encoder_drop_rate": 0.1,
        "encoder_input_dim": 896,
        "encoder_layer_config": "transformer",
        "encoder_output_dim": 896,
        "encoder_pre_norm_type": "ln",
        "encoder_upsample_rate": 9,
        "kv_cache_prefix_finetune": 0,
        "epochs": 100,
        "eps": 1e-08,
        "eps_decay": 0.8,
        "gpu_id": null,
        "gpu_num": 1,
        "grad_clip": 5,
        "grad_noise": false,
        "idim": 896,
        "init_lr": 0.0005,
        "lsm_weight": 0.0,
        "max_batch_size": 25,
        "max_duration": 256,
        "max_mem": 20000,
        "mtlalpha": 0.5,
        "n_iter_processes": 8,
        "noam_warmup_steps": 4000,
        "odim": 1024,
        "opt": "noamw",
        "rank": 0,
        "report_interval_iters": 100,
        "resume_trainer": false,
        "save_interval_iters": 2000,
        "seed": 19832,
        "sort_duration": true,
        "start_decay_epoch": 5,
        "stop_learning_rate": 1e-05,
        "sycn_batchnorm": false,
        "tensorboard_dir": null,
        "train_dtype": "bfloat16",
        "transformer_attention_dim": 896,
        "transformer_attention_dropout_rate": 0.1,
        "transformer_attention_heads": 14,
        "transformer_chunk_size": [
            1
        ],
        "transformer_concat_after": false,
        "transformer_dropout_rate": 0.1,
        "transformer_dynamic_chunks": false,
        "transformer_input_dim": 896,
        "transformer_input_layer": "linear",
        "transformer_left_chunks": [
            -1
        ],
        "transformer_linear_units": 4864,
        "transformer_normalize_before": true,
        "transformer_num_blocks": 4,
        "transformer_output_dim": 896,
        "transformer_pos_enc_class": "rel-enc",
        "transformer_positional_dropout_rate": 0.1,
        "transformer_positionwise_conv_kernel_size": 1,
        "transformer_positionwise_layer_type": "linear",
        "use_zero_redun_opt": false,
        "verbose": 0,
        "weight_decay": 0.05,
        "world_size": 1
    }
]