Print summary of all tests.
(self, total_time)
| 488 | print(f" ✅ Log saved: {log_file}") |
| 489 | |
| 490 | def print_summary(self, total_time): |
| 491 | """Print summary of all tests.""" |
| 492 | print(f"\n\n{'='*80}") |
| 493 | print(f"📊 TEST SUMMARY") |
| 494 | print(f"{'='*80}\n") |
| 495 | |
| 496 | # Sort results by perplexity (lower is better) |
| 497 | successful = [r for r in self.results if r['perplexity'] is not None] |
| 498 | failed = [r for r in self.results if r['perplexity'] is None] |
| 499 | |
| 500 | if successful: |
| 501 | # Extract numeric value from "mean±std" format for sorting |
| 502 | def get_ppl_value(result): |
| 503 | ppl = result['perplexity'] |
| 504 | if isinstance(ppl, str) and '±' in ppl: |
| 505 | return float(ppl.split('±')[0]) |
| 506 | elif isinstance(ppl, str): |
| 507 | try: |
| 508 | return float(ppl) |
| 509 | except ValueError: |
| 510 | return float('inf') |
| 511 | return ppl |
| 512 | |
| 513 | successful_sorted = sorted(successful, key=get_ppl_value) |
| 514 | |
| 515 | print(f"{'Dataset':<20} {'Perplexity':>20} {'Time (s)':>12} {'Status':<15}") |
| 516 | print(f"{'-'*80}") |
| 517 | |
| 518 | for result in successful_sorted: |
| 519 | ppl_str = str(result['perplexity']) if result['perplexity'] is not None else 'N/A' |
| 520 | print(f"{result['dataset']:<20} {ppl_str:>20} " |
| 521 | f"{result['time']:>12.2f} {result['status']:<15}") |
| 522 | |
| 523 | best_ppl = str(successful_sorted[0]['perplexity']) |
| 524 | print(f"\n🏆 Best result: {successful_sorted[0]['dataset']} " |
| 525 | f"(PPL: {best_ppl})") |
| 526 | |
| 527 | if failed: |
| 528 | print(f"\n❌ Failed tests ({len(failed)}):") |
| 529 | for result in failed: |
| 530 | print(f" - {result['dataset']}: {result['status']}") |
| 531 | |
| 532 | print(f"\n{'='*80}") |
| 533 | print(f"✅ Completed: {len(successful)}/{len(self.results)}") |
| 534 | print(f"⏱️ Total time: {total_time:.2f}s ({total_time/60:.2f} min)") |
| 535 | print(f"📁 Results saved in: {self.output_dir}") |
| 536 | print(f"{'='*80}\n") |
| 537 | |
| 538 | |
| 539 | def main(): |