{"id":20298058,"url":"https://github.com/sail-sg/Cheating-LLM-Benchmarks","last_synced_at":"2025-05-07T20:34:08.000Z","repository":{"id":259317072,"uuid":"870242119","full_name":"sail-sg/Cheating-LLM-Benchmarks","owner":"sail-sg","description":"[SafeGenAi @ NeurIPS 2024] Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates","archived":false,"fork":false,"pushed_at":"2024-10-23T21:52:33.000Z","size":1587,"stargazers_count":48,"open_issues_count":0,"forks_count":0,"subscribers_count":3,"default_branch":"main","last_synced_at":"2024-10-24T10:30:32.046Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Jupyter Notebook","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/sail-sg.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2024-10-09T17:29:46.000Z","updated_at":"2024-10-24T09:39:32.000Z","dependencies_parsed_at":"2024-10-24T10:39:43.012Z","dependency_job_id":null,"html_url":"https://github.com/sail-sg/Cheating-LLM-Benchmarks","commit_stats":null,"previous_names":["sail-sg/cheating-llm-benchmarks"],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sail-sg%2FCheating-LLM-Benchmarks","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sail-sg%2FCheating-LLM-Benchmarks/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sail-sg%2FCheating-LLM-Benchmarks/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/sail-sg%2FCheating-LLM-Benchmarks/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/sail-sg","download_url":"https://codeload.github.com/sail-sg/Cheating-LLM-Benchmarks/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":252953716,"owners_count":21830890,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2024-11-14T16:02:02.012Z","updated_at":"2025-05-07T20:34:07.988Z","avatar_url":"https://github.com/sail-sg.png","language":"Jupyter Notebook","funding_links":[],"categories":["A01_文本生成_文本对话"],"sub_categories":["大语言对话模型及数据"],"readme":"\u003ch1 align='center' style=\"text-align:center; font-weight:bold; font-size:2.0em;letter-spacing:2.0px;\"\u003e Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates \u003c/h1\u003e\n\n\u003cp align='left' style=\"text-align:left;font-size:1.2em;\"\u003e\n\u003cb\u003e\n    [\u003ca href=\"https://arxiv.org/abs/2410.07137\" target=\"_blank\" style=\"text-decoration: none;\"\u003earXiv\u003c/a\u003e] \n\u003c/b\u003e\n\u003c/p\u003e\n\n![img](./viz/leaderboard.jpeg)\n\n# Craft the null response\n\nRun [notebook_gpt4/gpt-4-1106-preview_vs_nil.ipynb](notebook_gpt4/gpt-4-1106-preview_vs_nil.ipynb) to get the null response augmented with the adversarial string. \n\n\n# Evaluation\n\n## Step 1: Prepare the submission\n\nRun [01_prepare_submission.ipynb](./01_prepare_submission.ipynb) to craft the null model submission.\n\n## Step 2: Evaluate the submission using alpaca-eval\n\nTo install the stable release of AlpacaEval 2.0, run\n\n```bash\npip install alpaca-eval\n```\n\nThen you can use it to evaluate the submission as follows:\n\n```bash\nexport OPENAI_API_KEY=\u003cyour_api_key\u003e # for more complex configs, e.g. using Azure or switching clients see client_configs/README.md \nalpaca_eval --model_outputs 'example/outputs.json' \n```\n\n## Step 3 (Optional): Re-evaluate the submission for further analysis\n\nRun [02_re_evaluate_submission.ipynb](./02_re_evaluate_submission.ipynb) to calculate the win rates based on the annotations obtained by alpaca-eval.\n\nFor example, you can get the following win rates using the [alpaca-eval annotations](./example/weighted_alpaca_eval_gpt4_turbo/annotations.json) of our null model.\n\n```\n{'win_rate': 76.91979180386511, \n'standard_error': 0.909010244966257, \n'n_wins': 676, \n'n_wins_base': 129, \n'n_draws': 0, \n'n_total': 805, \n'discrete_win_rate': 83.97515527950311,\n'length_controlled_winrate': 86.45780691307944, \n'lc_standard_error': 0.1418000511342794}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fsail-sg%2FCheating-LLM-Benchmarks","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fsail-sg%2FCheating-LLM-Benchmarks","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fsail-sg%2FCheating-LLM-Benchmarks/lists"}