{"data":{"kind":"file","path":"README.md","version_id":"zh0xlr2wp49blhjertu6jw8n","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1044,"modified_at":"2026-08-07T09:55:29.218000","content_hash":"7a5035c5a837b6344eee56eacb4f264da8adb4ee4530806748528ea42ce5cf25"},"entries":[],"content":"# python-coding-base\n\nHumanEval Python code generation environment (fork of `mohamed313/python-coding-eval`\nfor a code-generation RL experiment).\n\nThe model completes a Python function; the completion is run against the HumanEval\nunit tests in an isolated subprocess with a timeout and import-safety checks.\n\n## Variant\n\n- **python-coding-base** — original binary reward: 1.0 if all unit tests pass, else 0.0.\n- **python-coding-partial** — modified reward (small first change): 1.0 if all tests pass,\n  0.3 if the code defines a callable entry point but fails some tests, else 0.0.\n  Denser reward signal toward writing structurally valid code.\n\n## Usage\n\n```python\nfrom verifiers import load_environment\nenv = load_environment(\"python-coding-base\", num_examples=100, offset=0)\n```\n\n## Notes\n\n- Dataset: `openai/openai_humaneval` (split `test`, 164 problems).\n- `num_examples` and `offset` select a slice of the 164 problems, so training and\n  held-out eval can consume disjoint slices.\n- Source: https://github.com/PrimeIntellect-ai/prime\n","encoding":"utf-8","truncated":false,"total_bytes":1044},"status":null}