| { |
| "outputs": "evals/qwen3_14b_orca_refusal_release_grounded_outputs.jsonl", |
| "total": 54, |
| "passed": 54, |
| "pass_rate": 1.0, |
| "by_type": { |
| "source_packet": { |
| "passed": 50, |
| "total": 50 |
| }, |
| "hard_negative": { |
| "passed": 4, |
| "total": 4 |
| } |
| }, |
| "details": [ |
| { |
| "id": "grounded-release-eval-001", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-002", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-003", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-004", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-005", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-006", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-007", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-release-eval-008", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-009", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-release-eval-010", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-011", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-012", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-013", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-014", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-015", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-016", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-017", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-018", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-019", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-020", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-021", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-022", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-023", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-024", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-025", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-release-eval-026", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-027", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-release-eval-028", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-029", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-030", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-031", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-032", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-033", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-034", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-035", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-036", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S4" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-037", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-038", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-039", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-040", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-041", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-042", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-043", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-044", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-045", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-release-eval-046", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-047", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-048", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-049", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-release-eval-050", |
| "eval_type": "source_packet", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1", |
| "S2", |
| "S3", |
| "S4" |
| ], |
| "cited_source_ids": [ |
| "S1", |
| "S2", |
| "S3" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": false |
| }, |
| { |
| "id": "grounded-hard-001", |
| "eval_type": "hard_negative", |
| "passed": true, |
| "prompt_source_ids": [], |
| "cited_source_ids": [], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-hard-002", |
| "eval_type": "hard_negative", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-hard-003", |
| "eval_type": "hard_negative", |
| "passed": true, |
| "prompt_source_ids": [ |
| "S1" |
| ], |
| "cited_source_ids": [ |
| "S1" |
| ], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| }, |
| { |
| "id": "grounded-hard-004", |
| "eval_type": "hard_negative", |
| "passed": true, |
| "prompt_source_ids": [], |
| "cited_source_ids": [], |
| "invalid_citations": [], |
| "external_urls": [], |
| "has_uncertainty_language": true |
| } |
| ] |
| } |
|
|
