training_summary.json

{
  "model": "QueryAwareImportanceHead",
  "base_model": "mistralai/Mistral-7B-v0.3",
  "n_samples": 500,
  "epochs": 10,
  "batch_size": 2,
  "learning_rate": 0.0001,
  "best_loss": 0.6157153167724609,
  "checkpoint_best": "checkpoints/p3b_production/best.pt",
  "checkpoint_final": "checkpoints/p3b_production/final.pt"
}