{
  "base_model": "CohereLabs/cohere-transcribe-arabic-07-2026",
  "variant": "full",
  "method": "decoder_only_finetune",
  "target_column": "english",
  "dataset": "Ahmed1/cohere-asr-cs",
  "effective_batch_size": 32,
  "seed": 42,
  "bf16": true,
  "train_examples": 1900,
  "val_examples": 454,
  "metrics": {
    "train_runtime": 419.5181,
    "train_samples_per_second": 27.174,
    "train_steps_per_second": 0.858,
    "total_flos": 5.372639220298678e+19,
    "train_loss": 0.10716057336992688,
    "epoch": 6.0
  },
  "freeze": {
    "variant": "full",
    "decoder_module": "model.decoder",
    "decoder_layers_total": 8,
    "decoder_layers_trained": [
      0,
      1,
      2,
      3,
      4,
      5,
      6,
      7
    ],
    "train_decoder_io": true,
    "output_head": "proj_out",
    "io_trainable_params": 35935232,
    "layer_trainable_params": 134373376,
    "encoder_module": "model.encoder",
    "encoder_trainable_params": 0,
    "total_params": 2065647872,
    "trainable_params": 170308608,
    "trainable_pct": 8.245
  },
  "hyperparameters": {
    "learning_rate": 8e-05,
    "num_train_epochs": 6,
    "lr_scheduler_type": "cosine",
    "warmup_ratio": 0.03,
    "per_device_train_batch_size": 16,
    "gradient_accumulation_steps": 2,
    "optim": "adamw_bnb_8bit",
    "weight_decay": 0.01,
    "max_grad_norm": 1.0,
    "save_total_limit": 1
  }
}