{
    "training": {
        "wandb_name": "stemz_pre_suf_transfer",
        "wandb_project": "harmonai_train_vt",
        "tot_num_steps": 10000000,
        "batch_size": 2,
        "ckpt_every": 10000,
        "learning_rate": 5e-05,
        "lr_warmup": 2000,
        "lr_scheduler": "constant",
        "model_type": "prefix",
        "weight_decay": 0.001,
        "clip_grad_norm": 0.5,
        "betas": [
            0.9,
            0.999
        ],
        "seed_offset": 2,
        "compute_metrics": false,
        "use_ema": true,
        "activation_checkpointing": false,
        "compile": true,
        "preload_checkpoint": "/app/suno/checkpoints/2025-03-06_23-22-48_s4786/step_40000_infer.pt",
        "preload_optimizer": false,
        "check_model_args": false
    },
    "model": {
        "io_hz": 25,
        "io_channels": 128,
        "embed_dim": 2048,
        "depth": 32,
        "n_heads": 32,
        "qk_norm": true,
        "block_size": 500,
        "cond_semantic_n_vocab": 4001,
        "cond_semantic_len": 500,
        "cond_text_n_vocab": 60001,
        "cond_text_len": 1536,
        "ctx_len": 250,
        "infill_ctx_len": 500,
        "stem_ctx_len": 500
    },
    "data": {
        "shard_data": false,
        "shard_data_dir": "/mnt/localdisk/tmp",
        "allow_shard_reuse": false,
        "dataset_type": "stem_memmap",
        "dataset_dir": "/app/suno/data/diffusion_mix/dac_vae_tuned_25hz_stems_grouped",
        "train_vae_memmap_filename": "data_vae_tr.bin",
        "train_semantic_memmap_filename": "data_semantic_tr.bin",
        "train_metas_filename": "metas_tr.jsonl",
        "val_vae_memmap_filename": "data_vae_val.bin",
        "val_semantic_memmap_filename": "data_semantic_val.bin",
        "val_metas_filename": "metas_val.jsonl",
        "vae_dim": 128,
        "semantic_rate_hz": 25,
        "vae_scale_factor": 0.4,
        "codec_dir": "/app/suno/models/codecs",
        "foreign_weight": 2.0,
        "scale_vae_ctx": true,
        "noise_ctx": 1.0,
        "aligned_text_prob": 0.05,
        "always_pad_semantic": true,
        "max_multi_instruments": 10
    }
}