{"data":{"kind":"file","path":"README.md","version_id":"khvk2zon39hotdx8w2jzhnsm","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2193,"modified_at":"2026-03-14T06:56:02.299000","content_hash":"6e53d37890b97a07068cd8f5787d1d691773f9acce76e161461c07d1cf117ee1"},"entries":[],"content":"# Playgent Investment Banking Environment\n\nAn RL environment for realistic investment banking tasks.\n\n## Installation\n\n```bash\nprime env install playgent/playgent-investment-banking\n```\n\n## Dataset\n\nThe dataset is publicly available on Hugging Face: [playgent-hf/investment-banking](https://huggingface.co/datasets/playgent-hf/investment-banking)\n\n## Running an Evaluation\n\nSet your OpenAI API key for the LLM judge:\n\n```bash\nexport OPENAI_API_KEY=\"your-key\"\n```\n\n### Run a single task by ID\n\n```bash\nprime eval run playgent-investment-banking -r 1 \\\n  -m anthropic/claude-opus-4-5 \\\n  --env-args '{\"task_id\": \"task_xatm_ma_1_atm_loan_amortization\", \"output_dir\": \"./results\"}'\n```\n\nThe `output_dir` argument should be a local path where a zipped file of the **modified environment** will be. If you open this zip, you will see the world for the specific task you ran along with the answer file.\n\nThe `-r` argument refers to the number of rollouts. Rewards are averaged across rollouts.\n\n## Env Args\n\nPass these via `--env-args '{\"key\": \"value\"}'`:\n\n| Arg             | Default                 | Description                                                  |\n| --------------- | ----------------------- | ------------------------------------------------------------ |\n| `task_id`       | `null`                  | Run a specific task by ID                                    |\n| `output_dir`    | `null`                  | Directory to save snapshots and verifier results             |\n| `max_turns`     | `150`                   | Maximum agent turns per rollout                              |\n| `task_config`   | `null`                  | Path to a local JSON task config (alternative to HF dataset) |\n| `resum_enabled` | `true`                  | Enable context summarization when approaching context limit  |\n| `resum_model`   | `\"openai/gpt-4.1-mini\"` | LiteLLM model used for summarization                         |\n\n## Environment Variables\n\n| Variable            | Required For                       |\n| ------------------- | ---------------------------------- |\n| `ANTHROPIC_API_KEY` | Claude judge / summarization model |\n| `OPENAI_API_KEY`    | GPT judge / summarization model    |\n","encoding":"utf-8","truncated":false,"total_bytes":2193},"status":null}