| { | |
| "config_general": { | |
| "lighteval_sha": "?", | |
| "num_fewshot_seeds": 1, | |
| "max_samples": null, | |
| "job_id": 0, | |
| "start_time": 6052.722710683, | |
| "end_time": 6448.248247955, | |
| "total_evaluation_time_secondes": "395.5255372720003", | |
| "model_name": "Qwen/Qwen3-4B", | |
| "model_sha": "", | |
| "model_dtype": null, | |
| "model_size": null, | |
| "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": 0.5475268776886557, | |
| "truthfulqa_mc2_stderr": 0.015802591856386182 | |
| }, | |
| "all": { | |
| "truthfulqa_mc1": 0.3671970624235006, | |
| "truthfulqa_mc1_stderr": 0.01687480500145318, | |
| "truthfulqa_mc2": 0.5475268776886557, | |
| "truthfulqa_mc2_stderr": 0.015802591856386182 | |
| } | |
| }, | |
| "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": "13cc8c1ed56c3c97", | |
| "hash_cont_tokens": "2aa05ab785b97e1d" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 817, | |
| "padded": 0, | |
| "non_padded": 9996, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "aed1dfc67e53d0f2", | |
| "hash_full_prompts": "aed1dfc67e53d0f2", | |
| "hash_input_tokens": "e0d42156d66ed07f", | |
| "hash_cont_tokens": "bfa433470f6b6989" | |
| }, | |
| "truncated": 0, | |
| "non_truncated": 817, | |
| "padded": 0, | |
| "non_padded": 9996, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |