{"data":{"kind":"file","path":"README.md","version_id":"q2ed7l2nlp57s31at5zzy41d","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1733,"modified_at":"2025-09-20T06:10:27.174000","content_hash":"42fb0e15936f9dbd653800658fb65e6a403100a479727b5704de09be9014a14c"},"entries":[],"content":"# paraphrased-sentence-repeater\n\n### Overview\n- **Environment ID**: `paraphrased-sentence-repeater`\n- **Short description**: Environment that asks a model to return the sentence before or after a target sentence when provided with the paraphased version of the target sentence.\n- **Tags**: `paraphrased-sentence-repeater`, `train`, `eval`, `single-turn`, `long-context`\n\n### Datasets\n- **Primary dataset(s)**: `nreHieW/longwriter-sentences-paraphrased-binned`\n- **Source links**: https://huggingface.co/datasets/nreHieW/longwriter-sentences-paraphrased-binned\n- **Split sizes**: 1000 train\n\n### Task\n- **Type**: Single-turn\n- **Parser**: `None`\n- **Rubric overview**: The model is rewarded based on the similarity between the model's output and the target sentence as measured by the Longest Common Subsequence (LCS) ratio.\n\n### Quickstart\nRun an evaluation with default settings:\n\n```bash\nuv run vf-eval paraphrased-sentence-repeater\n```\n\nConfigure model and sampling:\n\n```bash\nuv run vf-eval paraphrased-sentence-repeater   -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\nThe environment supports the following arguments:\n\n| Arg | Type | Default | Description |\n| --- | ---- | ------- | ----------- |\n| `num_samples_per_bin` | int | `200` | Number of samples to use per bin |\n| `max_seq_len` | int | `128000` | Maximum sequence length |\n\n### Metrics\nThe rubric emits the LCS ratio between the model's output and the target sentence.\n\n| Metric | Meaning |\n| ------ | ------- |\n| `lcs_ratio` | LCS ratio between the model's output and the target sentence |\n\n","encoding":"utf-8","truncated":false,"total_bytes":1733},"status":null}