{
    "training": {
        "wandb_name": "1n_v3_flow_shared_pretrain_5E5_overlap_test",
        "wandb_project": "harmonai_train_vt",
        "tot_num_steps": 10000000,
        "batch_size": 1,
        "batch_reuse_factor": 4,
        "ckpt_every" : 10000,
        "learning_rate": 5e-05,
        "lr_warmup": 10000,
        "lr_scheduler": "constant",
        "model_type": "prefix",
        "weight_decay": 0.001,
        "clip_grad_norm": 0.5,
        "diffusion_objective": "rectified_flow",
        "noise_schedule": "log_snr",
        "betas": [
            0.9,
            0.999
        ],
        "seed_offset": 4,
        "compute_metrics": false,
        "use_ema": true,
        "activation_checkpointing": false,
        "compile": true,
        "preload_checkpoint": "/app2/suno/checkpoints/2025-11-10_20-02-05_s2469/last_ckpt.pt",
        "preload_optimizer": false,
        "check_model_args": false,
        "num_workers": 2,
        "profile_timing": true,
        "profile_print_every": 50
    },
    "model": {
        "io_hz": 25,
        "io_channels": 128,
        "embed_dim": 2048,
        "depth": 32,
        "n_heads": 32,
        "qk_norm": true,
        "block_size": 750,
        "cond_semantic_n_vocab": 4001,
        "cond_semantic_len": 750,
        "cond_text_n_vocab": 60001,
        "cond_text_len": 1536,
        "ctx_len": 750,
        "infill_ctx_len": 0,
        "shared_ctx": true,
        "use_actnorm": false,
        "actnorm_sync_init": false,
        "use_mmdit": false,
        "init_variant": "stability_v1",
        "use_rvq": false,
        "n_codebooks": 1
    },
    "data": {
        "dataset_type": "dynamic",
        "codec_filepath": "/app2/suno/data/christian/models/dac_vae_tuned_25hz.pth",
        "semantic_model_filepath": "/app2/suno/data/christian/models/mert_25.pt",
        "semantic_clusters_filepath": "/app2/suno/data/christian/models/mert_25_1x4k.npy",
        "train_metas_filepath": "/app2/suno/data/diffusion/v1/metas_diff_v4_val.jsonl",
        "val_metas_filepath": "/app2/suno/data/diffusion/v1/metas_diff_v4_val.jsonl",
        "audio_chunk_s": 30.02,
        "audio_ctx_s": 30.02,
        "audio_vox_s": 30.02,
        "vae_dim": 128,
        "semantic_rate_hz": 25,
        "vae_scale_factor": 0.4,
        "foreign_weight" : 1.0,
        "text_aligned_weight" : 3.0,
        "stem_weight" : 2.0,
        "scale_vae_ctx" : true,
        "noise_ctx" : 0.75,
        "use_text_aligned_prob": 0.25,
        "text_drop_prob": 0.1,
        "semantic_mask_prob": 0.1,
        "ctx_mask_prob": 0.2,
        "use_stem_prob" : 0.1,
        "use_vox_prob" : 0.9,
        "infill_prob": 0.1,
        "infill_min_ratio": 0.2,
        "infill_max_ratio": 0.8,
        "target_loudness_db": -16.0
    }
}