{"data":{"kind":"file","path":"README.md","version_id":"hn6jkot0qpmjfsla3ggva3qa","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1257,"modified_at":"2026-07-31T10:26:35.787000","content_hash":"91cd2f0af1068adc7347c2d1d1dd26c0e8052cce1103d966c5eae14215e65b13"},"entries":[],"content":"# Math Reasoning Environment\n\nA multi-metric math reasoning environment for training and evaluating LLMs on step-by-step mathematical problem solving.\n\n## Overview\n\nThis environment evaluates mathematical reasoning across three dimensions:\n\n1. **Answer Correctness** (weight: 0.6) - Does the final numerical answer match the ground truth?\n2. **Reasoning Quality** (weight: 0.25) - Does the response show clear step-by-step reasoning with math expressions?\n3. **Format Compliance** (weight: 0.15) - Does the response use the `####` marker for the final answer?\n\n## Dataset\n\nUses GSM8K (Grade School Math 8K) problems by default. Each problem requires multi-step arithmetic reasoning.\n\n## Usage\n\n```python\nimport verifiers as vf\n\n# Load with defaults (20 training examples)\nenv = vf.load_environment(\"math-reasoning\")\n\n# Custom configuration\nenv = vf.load_environment(\"math-reasoning\", num_examples=50, split=\"test\")\n```\n\n## Evaluation\n\n```bash\nprime eval run OWNER/math-reasoning -m Qwen/Qwen3.5-0.8B\n```\n\n## Rubric Functions\n\n- `answer_correctness`: Extracts and compares final numerical answers\n- `reasoning_quality`: Checks for step structure, math expressions, and clear conclusions\n- `format_compliance`: Verifies `####` marker usage\n\n## Version\n\n0.1.2\n","encoding":"utf-8","truncated":false,"total_bytes":1257},"status":null}