{"data":{"kind":"file","path":"README.md","version_id":"imabilqsycr6o953rc9wu3dm","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2154,"modified_at":"2026-01-18T18:02:35.182000","content_hash":"ffa345eb92f0a52ecee5cc4f9168e92bb2db0cb42a113f9addb49560331e66e0"},"entries":[],"content":"# datbench-env\n\n### Overview\n- **Environment ID**: `datbench-env`\n- **Short description**: Verifiers wrapper for Datology's [DatBench](https://github.com/datologyai/DatBench) eval library\n- **Tags**: multimodal, vision, vqa, eval\n\n### Datasets\n- **Primary dataset(s)**: DatBench curated dataset\n- **Source links**: https://huggingface.co/datasets/DatologyAI/DatBench\n- **Split sizes**: Multiple capability subsets spanning 33 widely used VLM benchmarks.\n\n### Task\n- **Type**: `vf.SingleTurnEnv`\n- **Parser**: `vf.MaybeThinkParser`.\n- **Rubric overview**: Scoring harness for DatBench datasets.\n\n### Implementation Notes\n- **External dependencies**: `cc-ocr-doc_parsing` and `ocrbench-v2` tasks from `document` require `libxml2` and `libxslt` for more accurate scoring. [Installation](https://lxml.de/installation.html)\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nuv run vf-eval datbench-env\n```\n\nConfigure model and sampling:\n\n```bash\nuv run vf-eval datbench-env   -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\n\n| Arg | Type | Default | Description |\n| --- | ---- | ------- | ----------- |\n| `dataset_name` | str | `\"DatologyAI/DatBench\"` | Dataset name from Hugging Face |\n| `capability` | str | `\"math\"` | Dataset capability |\n| `subset` | str | `None` | Source dataset/benchmark |\n| `dataset_split` | str | `\"test\"` | Dataset split |\n| `max_examples` | int | `-1` | Limit on dataset size (use -1 for all). Useful for pre-processing only a subset of the dataset. |\n| `judge_model` | str | `\"gpt-4o\"` | Judge model |\n| `judge_base_url` | str | `\"https://api.openai.com/v1\"` | Judge base URL |\n| `judge_api_key_var` | str | `\"OPENAI_API_KEY\"` | Judge API key variable |\n\n### Metrics\n\n| Metric | Meaning |\n| ------ | ------- |\n| `reward` | Main scalar reward (weighted sum of criteria) |\n| `direct_reward` | Score from DatBench harness when `eval_mode` is `direct` |\n| `judge_reward` | LLM-as-judge score from DatBench harness when `eval_mode` is `judge` |","encoding":"utf-8","truncated":false,"total_bytes":2154},"status":null}