LeapyDeapy/test_run_tiny_kb_leap2

ℹ️ Model information

• Base model: KBLab/kb-whisper-tiny
• Training data: LeapyDeapy/handkirurgi-v1
• Training config: {'use_cuda': True, 'cuda_device': 0, 'batch_size': 8, 'eval_batch_size': 8, 'gradient_accumulation_steps': 1, 'learning_rate': '3e-5', 'warmup_steps': 1, 'epochs': 1, 'evaluation_strategy': 'epoch', 'save_strategy': 'epoch', 'save_steps': 1, 'logging_steps': 1, 'generation_max_length': 128, 'fp16': True, 'num_proc': 1, 'remove_unused_columns': False, 'label_names': ['labels'], 'predict_with_generate': True, 'load_best_model_at_end': True, 'metric_for_best_model': 'wer', 'greater_is_better': False}

---

test_run_tiny_kb_leap2

This is a merged Whisper model fine-tuned with PEFT adapters.

🧪 Evaluation Metrics

Metric	Value
Eval WER	63.96
Eval Loss	3.6525
Epoch	1.0
Runtime (s)	33.9135

📁 Files Included

📌 Notes

This model was trained on runs\test_run_tiny_kb_leap2 and auto-uploaded with metrics for traceability.