| { |
| "architecture": "aiva-rlcd-multihead", |
| "base_model": "answerdotai/ModernBERT-base", |
| "base_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8", |
| "dataset": "s-ttp/aiva-rlcd-labels-v2", |
| "dataset_revision": "7c3c0971267a011320d487af0c30bea6c4e18f7f", |
| "pooling": "mean", |
| "max_length": 1024, |
| "heads": { |
| "task_category": [ |
| "qa", |
| "summarization", |
| "coding", |
| "rag", |
| "analysis", |
| "generation", |
| "agentic" |
| ], |
| "sensitivity": [ |
| "public", |
| "internal", |
| "confidential", |
| "restricted", |
| "sovereign-only" |
| ], |
| "complexity": [ |
| "simple", |
| "moderate", |
| "complex", |
| "advanced" |
| ], |
| "context_need": [ |
| "sufficient", |
| "needs-retrieval", |
| "needs-packaging" |
| ] |
| }, |
| "temperatures": { |
| "task_category": 1.0844, |
| "sensitivity": 0.9798, |
| "complexity": 1.4548, |
| "context_need": 1.0657 |
| }, |
| "input_format": "messages joined as '[role] content' lines (aiva_ml.rlcd.labels.render)", |
| "hyperparameters": { |
| "base_model": "answerdotai/ModernBERT-base", |
| "dataset": "s-ttp/aiva-rlcd-labels-v2", |
| "dataset_revision": "7c3c0971267a011320d487af0c30bea6c4e18f7f", |
| "data_dir": "combined", |
| "balance_sensitivity": 0.5, |
| "bucket_by_length": true, |
| "prior": "public=0.1458,internal=0.5272,confidential=0.2970,restricted=0.0236,sovereign-only=0.0064", |
| "output_repo": "s-ttp/aiva-rlcd-modernbert-v2-staging", |
| "max_length": 1024, |
| "epochs": 3.0, |
| "learning_rate": 5e-05, |
| "batch_size": 32, |
| "max_train_samples": null, |
| "max_eval_samples": null, |
| "no_push": false |
| }, |
| "train_seconds": 641, |
| "test_report": { |
| "task_category": { |
| "classes": [ |
| "qa", |
| "summarization", |
| "coding", |
| "rag", |
| "analysis", |
| "generation", |
| "agentic" |
| ], |
| "test_rows": 3727, |
| "accuracy": 0.9911, |
| "macro_f1": 0.986, |
| "ece_before": 0.0061, |
| "ece_after": 0.0054, |
| "temperature": 1.0758, |
| "provenance": "measured" |
| }, |
| "sensitivity": { |
| "classes": [ |
| "public", |
| "internal", |
| "confidential", |
| "restricted", |
| "sovereign-only" |
| ], |
| "test_rows": 3727, |
| "accuracy": 0.9799, |
| "macro_f1": 0.7854, |
| "ece_before": 0.0115, |
| "ece_after": 0.0117, |
| "temperature": 0.9782, |
| "provenance": "measured" |
| }, |
| "complexity": { |
| "classes": [ |
| "simple", |
| "moderate", |
| "complex", |
| "advanced" |
| ], |
| "test_rows": 3727, |
| "accuracy": 0.8924, |
| "macro_f1": 0.8699, |
| "ece_before": 0.048, |
| "ece_after": 0.024, |
| "temperature": 1.4539, |
| "provenance": "measured" |
| }, |
| "context_need": { |
| "classes": [ |
| "sufficient", |
| "needs-retrieval", |
| "needs-packaging" |
| ], |
| "test_rows": 3727, |
| "accuracy": 0.971, |
| "macro_f1": 0.9719, |
| "ece_before": 0.0178, |
| "ece_after": 0.017, |
| "temperature": 1.0647, |
| "provenance": "measured" |
| } |
| }, |
| "calibration": { |
| "temperatures_fitted_on": "s-ttp/aiva-rlcd-labels-v2 natural/natural-validation.jsonl (2,018 twice-agreed rows), CPU float32 logits as the gateway computes them, NLL weighted to the realistic sensitivity prior", |
| "realistic_prior": { |
| "public": 0.1458, |
| "internal": 0.5272, |
| "confidential": 0.297, |
| "restricted": 0.0236, |
| "sovereign-only": 0.0064 |
| }, |
| "training_job_temperatures": { |
| "task_category": 1.0758048295974731, |
| "sensitivity": 0.9781951904296875, |
| "complexity": 1.4539148807525635, |
| "context_need": 1.0646779537200928 |
| } |
| }, |
| "evaluation": { |
| "report": "ml/reports/rlcd-v2-gates-run1.md (AiVa repository)", |
| "eval_run": "s-ttp/aiva-eval-runs runs/rlcd-v2-eval-run1", |
| "frozen_test": "s-ttp/aiva-rlcd-labels-v2@773a37d9c06934b46c52eb8a06314918332ee826 natural/natural-test.jsonl", |
| "gates_passed": "33 of 33 at max_length 1024", |
| "decision": "every gate passes at 1024; 512 does not", |
| "provenance": "measured" |
| }, |
| "training_dataset": { |
| "repo": "s-ttp/aiva-rlcd-labels-v2", |
| "revision": "7c3c0971267a011320d487af0c30bea6c4e18f7f", |
| "data_dir": "combined" |
| } |
| } |
