| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "max_samples": null, | |
| "job_id": 0, | |
| "start_time": 1697.241612831, | |
| "end_time": 2750.279500025, | |
| "total_evaluation_time_secondes": "1053.0378871939997", | |
| "model_name": "Qwen/Qwen3-4B", | |
| "model_sha": "82d62bb073771e7a1ea59435f548908540217d1f", | |
| "model_dtype": null, | |
| "model_size": "14.98 GB", | |
| "generation_parameters": { | |
| "early_stopping": null, | |
| "repetition_penalty": null, | |
| "frequency_penalty": null, | |
| "length_penalty": null, | |
| "presence_penalty": null, | |
| "max_new_tokens": null, | |
| "min_new_tokens": null, | |
| "seed": null, | |
| "stop_tokens": null, | |
| "temperature": null, | |
| "top_k": null, | |
| "min_p": null, | |
| "top_p": null, | |
| "truncate_prompt": null, | |
| "response_format": null | |
| } | |
| }, | |
| "results": { | |
| "leaderboard|truthfulqa:mc|0": { | |
| "truthfulqa_mc1": 0.3671970624235006, | |
| "truthfulqa_mc1_stderr": 0.01687480500145318, | |
| "truthfulqa_mc2": 0.5481532716523813, | |
| "truthfulqa_mc2_stderr": 0.015802558799916978 | |
| }, | |
| "all": { | |
| "truthfulqa_mc1": 0.3671970624235006, | |
| "truthfulqa_mc1_stderr": 0.01687480500145318, | |
| "truthfulqa_mc2": 0.5481532716523813, | |
| "truthfulqa_mc2_stderr": 0.015802558799916978 | |
| } | |
| }, | |
| "versions": { | |
| "leaderboard|truthfulqa:mc|0": 0 | |
| }, | |
| "config_tasks": { | |
| "leaderboard|truthfulqa:mc": { | |
| "name": "truthfulqa:mc", | |
| "prompt_function": "truthful_qa_multiple_choice", | |
| "hf_repo": "truthful_qa", | |
| "hf_subset": "multiple_choice", | |
| "metric": [ | |
| { | |
| "metric_name": [ | |
| "truthfulqa_mc1", | |
| "truthfulqa_mc2" | |
| ], | |
| "higher_is_better": { | |
| "truthfulqa_mc1": true, | |
| "truthfulqa_mc2": true | |
| }, | |
| "category": "8", | |
| "use_case": "1", | |
| "sample_level_fn": "truthfulqa_mc_metrics", | |
| "corpus_level_fn": { | |
| "truthfulqa_mc1": "mean", | |
| "truthfulqa_mc2": "mean" | |
| } | |
| } | |
| ], | |
| "hf_revision": null, | |
| "hf_filter": null, | |
| "hf_avail_splits": [ | |
| "validation" | |
| ], | |
| "trust_dataset": true, | |
| "evaluation_splits": [ | |
| "validation" | |
| ], | |
| "few_shots_split": null, | |
| "few_shots_select": null, | |
| "generation_size": -1, | |
| "generation_grammar": null, | |
| "stop_sequence": [ | |
| "\n" | |
| ], | |
| "num_samples": null, | |
| "suite": [ | |
| "leaderboard" | |
| ], | |
| "original_num_docs": 817, | |
| "effective_num_docs": 817, | |
| "must_remove_duplicate_docs": false, | |
| "version": 0 | |
| } | |
| }, | |
| "summary_tasks": { | |
| "leaderboard|truthfulqa:mc|0": { | |
| "hashes": { | |
| "hash_examples": "36a6d90e75d92d4a", | |
| "hash_full_prompts": "36a6d90e75d92d4a", | |
| "hash_input_tokens": "38466dd693501854", | |
| "hash_cont_tokens": "cceeffa697353554" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 817, | |
| "padded": 9186, | |
| "non_padded": 810, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "aed1dfc67e53d0f2", | |
| "hash_full_prompts": "aed1dfc67e53d0f2", | |
| "hash_input_tokens": "48256f08799e5a5e", | |
| "hash_cont_tokens": "ed734856e6e7940f" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 817, | |
| "padded": 9186, | |
| "non_padded": 810, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |