{"data":{"kind":"file","path":"README.md","version_id":"b1lj49pdmbipbqozs5gaks3n","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1414,"modified_at":"2026-02-14T04:09:41.481000","content_hash":"4c458afcda805d661e67576a7b3638e6920475956c9f6d9f4da0c718b3430c5d"},"entries":[],"content":"# rlm-wrapper\n\n### Overview\n- **Environment ID**: `rlm-wrapper`\n- **Short description**: Wrapper that adapts any Verifiers env to the RLM interface.\n- **Tags**: wrapper, adapter, rlm, train, eval\n\n### Datasets\n- **Primary dataset(s)**: The dataset from the base environment.\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nprime eval run rlm-wrapper\n```\n\nConfigure model and sampling:\n\n```bash\nprime eval run rlm-wrapper   -m gpt-4.1-mini   -n 20 -r 3 -t 1024 -T 0.7   -a '{\"base_env_id\": \"primeintellect/alphabet-sort\", \"base_env_args\": {\"min_turns\": 3}, \"repl_language\": \"bash\"}' # Use alphabet-sort with min_turns=3 and a bash REPL\n```\n\nNotes:\n- Use `-a` / `--env-args` to environment-specific configuration as a JSON object.\n- `repl_language` currently defaults to `\"python\"`.\n\n### Environment Arguments\n\n| Arg | Type | Default | Description |\n| --- | ---- | ------- | ----------- |\n| `base_env_id` | str | `\"alphabet-sort\"` | Base environment ID |\n| `base_env_args` | dict | `{}` | Base environment arguments |\n| `**kwargs` | | | Any additional arguments are passed to the `RLMEnv` constructor; See [RLMEnv](https://github.com/PrimeIntellect-ai/verifiers/blob/c0508421e9550338706dd99d3ca3961ba4a29ace/verifiers/envs/experimental/rlm_env.py#L1927). |\n\n### Metrics\n\nThe metrics and rewards emitted by the base environment are passed through and aggregated with those from the `RLMEnv` wrapper.","encoding":"utf-8","truncated":false,"total_bytes":1414},"status":null}