training_run_config.json

{
  "args": {
    "features": "analysis_outputs/nvmos_precomputed_wavlm_spear_all_layers_seq_20260603/audio_spear_l9_none.pt",
    "text_features": "analysis_outputs/nvmos_precomputed_wavlm_spear_all_layers_seq_20260603/text_xlmr_large_tagctx.pt",
    "query_mode": "tag_ctx",
    "score_xlsx": "docs/nvmos/expert_score/final/LI_scoring_sheet_6_12.xlsx",
    "mapping_jsonl": "docs/nvmos/audio_by_nv_for_eval_numbered_mapping.jsonl",
    "manifest_json": "analysis_outputs/nvmos_precomputed_wavlm_spear_all_layers_seq_20260603/manifest.json",
    "output_dir": "analysis_outputs/nvmos_multinv_li612_textquery_wavlm_spear_all_layers_e10_20260613/spear_l9",
    "epochs": 10,
    "batch_size": 8,
    "lr": 0.0001,
    "weight_decay": 0.01,
    "hidden_dim": 256,
    "cross_layers": 2,
    "attention_heads": 8,
    "ffn_dim": 1024,
    "dropout": 0.1,
    "grad_clip": 1.0,
    "num_workers": 0,
    "seed": 1234,
    "device": "cuda"
  },
  "align_summary": {
    "score_rows": 7784,
    "aligned_rows": 7784,
    "missing_mapping_rows": 0,
    "missing_manifest_rows": 0,
    "tag_mismatch_rows": 0,
    "text_mismatch_rows": 1,
    "split_counts": {
      "valid": 389,
      "train": 7006,
      "test": 389
    },
    "tag_counts": {
      "ahem": 594,
      "chuckle": 275,
      "cough": 684,
      "cry": 594,
      "exhale": 215,
      "hiss": 549,
      "hum": 549,
      "inhale": 220,
      "laugh": 684,
      "moan": 542,
      "pant": 594,
      "sigh": 594,
      "smack": 504,
      "sneeze": 437,
      "sniffle": 594,
      "snore": 155
    }
  },
  "sizes": {
    "train": 7006,
    "valid": 389,
    "test": 389
  },
  "dropped": {
    "train": 0,
    "valid": 0,
    "test": 0
  },
  "audio_dim": 1024,
  "text_dim": 1024,
  "trainable_params": 2172929
}