shoal 检验
play
详情
+54 −0 scienceslack.json 新增 查看文件
| @@ -0,0 +1,54 @@ | ||
| 1 | +{ | |
| 2 | + "description": "ARC Prize 2026's interactive games, on the official toolkit, offline: no instructions, clear the levels in as few actions as a person. Bring an ARC-AGI-3 agent.", | |
| 3 | + "image": "python:3.12-slim", | |
| 4 | + "cover": { | |
| 5 | + "file": "cover.png" | |
| 6 | + }, | |
| 7 | + "game": { | |
| 8 | + "shelf": "research", | |
| 9 | + "play": [ | |
| 10 | + "casual" | |
| 11 | + ], | |
| 12 | + "mode": "both", | |
| 13 | + "players": "1", | |
| 14 | + "upstream": { | |
| 15 | + "name": "ARC-AGI-3 toolkit", | |
| 16 | + "url": "https://github.com/arcprize/ARC-AGI-3-Agents", | |
| 17 | + "license": "MIT", | |
| 18 | + "version": "arc-agi 0.9.9" | |
| 19 | + }, | |
| 20 | + "viewer": "viewer" | |
| 21 | + }, | |
| 22 | + "benchmark": { | |
| 23 | + "task": "ARC-AGI-3", | |
| 24 | + "metric": { | |
| 25 | + "name": "RHAE (%)", | |
| 26 | + "better": "higher" | |
| 27 | + }, | |
| 28 | + "subtasks": "games", | |
| 29 | + "aggregate": "mean", | |
| 30 | + "baseline": { | |
| 31 | + "name": "Random agent", | |
| 32 | + "score": 0.0, | |
| 33 | + "subtasks": { | |
| 34 | + "ft09": 0.0, | |
| 35 | + "lp85": 0.0, | |
| 36 | + "ls20": 0.0, | |
| 37 | + "sp80": 0.0, | |
| 38 | + "vc33": 0.0 | |
| 39 | + } | |
| 40 | + }, | |
| 41 | + "guardrails": [ | |
| 42 | + "At most 400 actions a game", | |
| 43 | + "The agent plays through the toolkit against the referee's server and never sees a game's source", | |
| 44 | + "Scored by ARC's RHAE from the server's own recording" | |
| 45 | + ], | |
| 46 | + "version": 1 | |
| 47 | + }, | |
| 48 | + "checks": { | |
| 49 | + "play": { | |
| 50 | + "path": "referee/check.sbatch", | |
| 51 | + "metrics": "metrics.json" | |
| 52 | + } | |
| 53 | + } | |
| 54 | +} | |