{"data":{"kind":"file","path":"README.md","version_id":"pjlmdy3snqbk7h1hx6tobu0t","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1988,"modified_at":"2025-09-26T07:57:43.092000","content_hash":"39eb831d0a7e28e929bddd8011bc83ebcec3b482658c55afd9467f77bbd543c7"},"entries":[],"content":"# gdpval\n\n### Overview\n- **Environment ID**: `gdpval`\n- **Short description**: GDPval-style benchmark environment for evaluating LLMs on realistic professional deliverables with an LLM-as-a-judge rubric.\n- **Tags**: evaluation, benchmark, gdpval, llm-as-a-judge\n\n### Datasets\n- **Primary dataset(s)**: openai/gdpval – 220 gold tasks spanning 44 occupations across 9 major sectors. Each task specifies a prompt and sometimes reference files.\n- **Source links**: [GDPval overview](https://openai.com/index/gdpval/), [dataset on Hugging Face](https://huggingface.co/datasets/openai/gdpval)\n- **Split sizes**: train = 220 examples (gold set)\n\n### Task\n- **Type**: single-turn (each task is one prompt → one deliverable)\n- **Parser**: no strict parsing; free-form completions graded by rubric\n- **Rubric overview**: Uses JudgeRubric with a custom grading prompt that scores outputs on:\n  - Task fulfillment & correctness\n  - Factuality and use of references\n  - Structure, formatting, professionalism\n  - Safety & compliance\n  - Reasoning quality\n  \n  The rubric returns a scalar reward ∈ [0.0, 1.0].\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nuv run vf-eval gdpval\n```\n\nConfigure model and sampling:\n\n```bash\nuv run vf-eval gdpval   -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| `foo` | str | `\"bar\"` | What this controls |\n| `max_examples` | int | `-1` | Limit on dataset size (use -1 for all) |\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` | Exact match on target answer |\n\n","encoding":"utf-8","truncated":false,"total_bytes":1988},"status":null}