Views
No views yet
Laplaces-Red-Devils/fol-v02-origin-qwen2.5-3premises_fol trong completion).experiment_artifacts/ trên repo này.train_metrics.json)fol_eval_metrics.json)| Split | Exact match | Đúng / Tổng |
|---|---|---|
| train | 0.2 | 4 / 20 |
| dev | 0 | 0 / 20 |
| test | 0 | 0 / 20 |
experiment_log.json)Qwen/Qwen2.5-3B-Instruct| Trước FT | Sau FT | |
|---|---|---|
| test exact match (full) | — | {'exact_match_count': 0, 'total': 80, 'exact_match_rate': 0.0} |
| test mean NLL completion | — | {'mean_nll_completion': 0.20448867972008883, 'n_examples': 80, 'split_size': 80} |
fol_inference_latency.json)test — n: 30experiment_artifacts/:configs/fol_model.yaml), fol_eval_* sau eval greedy, fol_inference_latency_benchmark nếu bật benchmark.Trainer.FOL_* (xem FolSFTConfig trong services/config_fol.py).Laplaces-Red-Devils/fol-v02-origin-qwen2.5-3 rồi chỉ greedy trên N mẫu test ngẫu nhiên; bảng dưới là exact-match trên đúng N mẫu đó. JSON đủ: experiment_artifacts/fol_hub_reload_eval.json.| Split | Accuracy | Đúng / Tổng |
|---|---|---|
| test | 0 | 0 / 20 |
split_index=14){"premises_fol": ["∀x(HasGPA(x))", "∀x(¬AttendsLectures(x) → ¬ImprovesGPA(x))", "∀x(SubmitsAssignments(x) → ImprovesGPA(x))", "∃x(HonorRoll(x))", "∀x(HighGPA(x) → EligibleForScholarship(x))", "∀x(HighGPA(x) → HonorRoll(x))", "∀x((HighGPA(x) …{"premises_fol": ["∀x (HasGPA(x))", "∀x (¬AttendsLectures(x) → ¬ImprovesGPA(x))", "∀x (SubmitsAssignmentsOnTime(x) → ImprovesGPA(x))", "∃x (OnHonorRoll(x))", "∀x (HighGPA(x) → EligibleForScholarship(x))", "∀x (HighGPA(x) → OnHonorRoll(x))", …split_index=3){"premises_fol": ["∀x (∀h ((clinical_hours(x, h) ∧ h ≥ 500) → advanced_practice(x)))", "clinical_hours(john, 600)", "registered_nurse(john)", "∀x ((registered_nurse(x) ∧ advanced_practice(x)) → can_prescribe(x))"]} …{"premises_fol": ["∀x (completed_500_hours(x) → advanced_practice_certification(x))", "completed_500_hours(John)", "registered_nurse(John)", "∀x (advanced_practice_certification(x) → authorized_prescribe_medication(x))"]} …