{"data":{"kind":"file","path":"README.md","version_id":"lbp60xchrrz552ll4v0stsos","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1829,"modified_at":"2025-12-01T20:10:14.570000","content_hash":"a08753cac26bfc4c26981fc663098c012e8b9c628fda4c0b41a92ab35b605d7b"},"entries":[],"content":"# xbench-scienceqa\n\n### Overview\n- **Environment ID**: `xbench-scienceqa`\n- **Short description**: A science question answering environment for evaluating scientific reasoning and problem-solving capabilities.\n- **Tags**: science, qa, chinese, llm-as-judge, eval\n\n### Datasets\n- **Primary dataset(s)**: ScienceQA.csv (commit `51ee5cf`)\n- **Source links**: https://github.com/xbench-ai/xbench-evals/\n- **Split sizes**: 100 examples for eval\n\n### Task\n- **Type**: SingleTurnEnv\n- **Parser**: `vf.Parser`\n- **Rubric overview**: JudgeRubric with correctness criteria. The main reward has a exact-match short-circuit, and falls back to a judge.\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nuv run vf-eval xbench-scienceqa\n```\n\nConfigure model and sampling:\n\n```bash\nuv run vf-eval xbench-scienceqa   -m gpt-4.1-mini   -n 20 -r 3 -t 1024 -T 0.7   -a '{\"key\": \"value\"}'  # env-specific args as JSON\n```\n\nNotes:\n- Use `-a` / `--env-args` to pass environment-specific configuration as a JSON object.\n\n### Environment Arguments\nDocument any supported environment arguments and their meaning. Example:\n\n| Arg | Type | Default | Description |\n| --- | ---- | ------- | ----------- |\n| `judge_model` | str | `\"gemini-flash-lite-latest\"` | Judge model to use |\n| `judge_base_url` | str | `\"https://generativelanguage.googleapis.com/v1beta/openai/\"` | Judge base URL |\n| `judge_api_key_var` | str | `\"GEMINI_API_KEY\"` | Judge API key variable |\n| `dataset_variation` | str | `\"ScienceQA.csv\"` | Dataset variation to use |\n| `max_turns` | int | `40` | Maximum number of turns |\n\n### Metrics\nSummarize key metrics your rubric emits and how they’re interpreted.\n\n| Metric | Meaning |\n| ------ | ------- |\n| `reward` | Main scalar reward (weighted sum of criteria) |\n| `accuracy` | Judge-provided accuracy on target answer |\n\n","encoding":"utf-8","truncated":false,"total_bytes":1829},"status":null}