{"data":{"kind":"file","path":"README.md","version_id":"ok6zjhi6pvv3u5w9yd08k5j3","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1718,"modified_at":"2025-10-04T05:33:12","content_hash":"7105f7c652fac209e3096ea952452d56ca618eeeed7ed29db4f387c7ae017ded"},"entries":[],"content":"# MulSeT\n\nThis environment is inspired by work done in this [paper](https://www.alphaxiv.org/abs/2509.02359) and their [code](https://github.com/WanyueZhang-ai/spatial-understanding). Their original code is strictly based on three architectures passed via embeddings.\n\n### Overview\n- **Environment ID**: `MulSeT`\n- **Short description**: Passes images and ask questions based on `Direction`, `Distance`, and `Mask` categories.\n- **Tags**: VLM, Image, Depth Analysis, Direction, Mask\n\n### Datasets\n- **Primary dataset(s)**: MulSet dataset introduced in the paper.\n- **Source links**: [HF dataset](https://huggingface.co/datasets/WanyueZhang/MulSeT)\n\n### Task\n- **Type**: single-turn\n- **Rubric overview**: If the response matches the correct answer.\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nuv run vf-eval MulSeT\n```\n\nConfigure model and sampling:\n\n```bash\nuv run vf-eval MulSeT \\ -k <API_KEY> -b <URL> -m <MODEL_NAME> -n 10 -r 1 -s \\\n  -a '{\"num_samples\": 10, \"categories\": \"dist,dir\", \"seed\": 4}'\n```\n\nNotes:\n- Use `-a` / `--env-args` to pass environment-specific configuration as a JSON object.\n- Could run multiple evals for each category.\n\n### Environment Arguments\nDocument any supported environment arguments and their meaning. Example:\n\n| Arg | Type | Default | Description |\n| --- | ---- | ------- | ----------- |\n| `num_samples` | int | 10 | Number of samples to load |\n| `categories` | str | `dist` | Different categories passed with `,` as delimiter; possible categories: **dir**, **dist**, and **mask** |\n\n### Metrics\nSummarize key metrics your rubric emits and how they’re interpreted.\n\n| Metric | Meaning |\n| ------ | ------- |\n| `reward` | If the answer matches the option |\n\n","encoding":"utf-8","truncated":false,"total_bytes":1718},"status":null}