Datasetpublic

ByteDance-Seed/EdgeBench

Overview EdgeBench is a benchmark of 134 real-world tasks for evaluating how autonomous AI agents learn from real-world environments. Instead of measuring one-shot performance, EdgeBench places agents in executable task environments with realistic, multi-level feedback and lets them iterate for 12+ hours per task — tracking the full trajectory of improvement, not just the final score. We publicly release 51 tasks… See the full description on the dataset page: https://huggingface.co/datasets/ByteDance-Seed/EdgeBench.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

84likes7.5kdownloads

roottrinity_text_adventure.json

trinity_text_adventure.json30 linesDownload Raw Back to root

1{2  "task_id": "trinity_text_adventure",3  "name": "Trinity Text Adventure",4  "category": "Interactive Games & Simulators",5  "base_image": "python310",6  "platform": "linux/amd64",7  "internet": false,8  "game_mode": true,9  "cwd": "/home/jericho_agent",10  "submit_paths": [],11  "work": {12    "image_tag": "61ded5caecc2",13    "specs_dir": "/home/jericho_agent",14    "agent_query": "## Trinity — Jericho Text Adventure\n\nPlay the interactive fiction game *Trinity* by sending commands via an HTTP API and maximize your score.\n"15  },16  "judge": {17    "image_tag": "fcaad5fa630f",18    "eval_cmd": "",19    "eval_timeout": 600,20    "parser": "",21    "selection": "score_first",22    "game_server_cmd": "python /tmp/game_server_app.py --rom /home/roms/trinity.z4",23    "rescale": {24      "kind": "linear",25      "lower": 0.0,26      "upper": 100.027    }28  }29}30