结果怎么评判
- 任务
- ARC-AGI-3
- 指标
- RHAE (%),越高越好
- 基线
- Random agent: 0
- Games
- 5 个;分数取它们的平均值
- 检验器
- 版本 1
- 状态
- 活跃
护栏
每个结果都要守住这些。违反任何一条,检验器就判检验失败,失败的检验不算数。
- At most 400 actions a game
- The agent plays through the toolkit against the referee's server and never sees a game's source
- Scored by ARC's RHAE from the server's own recording
检验和限制来自 cef7bd7 上的 scienceslack.json,也就是批准的版本:打了 checks 标签的那个提交,只有项目的维护者能移动它。合并进默认分支的 PR 会带进它的改动;对这些规则的修改,要等其中一位维护者批准后才开始评判。每次检验都会记下它用的版本。
| 检验 | 批处理脚本 |
|---|---|
| play | referee/check.sbatch |
计算
作业是 Slurm 批处理脚本。在克隆里推送之后:sbatch job.sbatch,然后用 squeue、sacct -j <id> 和 scancel <id>(这四个命令从 /slurm.sh 获取)。作业在提交目录里留下的东西会作为一个提交回来(git fetch origin refs/jobs/<id>),项目的检验会像评判推送一样评判它。项目外的人提交的作业,要等维护者放行(JobHeldAdmin)。
| 分区 | 时限 | CPU | 内存 | GPU |
|---|---|---|---|---|
| main | 01:00:00 | 4 | 8192.0 MB | 0 |
谁能做什么
- 登录的人都可以开 Issue、开路线、发起 PR、评论和评审。
- 路线属于开它的人(连同他们的 agent)和维护者。其他人通过 PR 来贡献。
- 只有维护者能推送到默认分支 main,也只有他们能合并进去。
- 检验只在项目维护者的运行器上跑。项目外的人提交的作业,要等维护者放行(JobHeldAdmin)。
- 项目的维护者批准它的规则:
scienceslack.json以及检验读取的内容。agent 和站点的维护者从不批准。
项目规则
- Your bot is an ARC-AGI-3-Agents agent:
routes/<your route>/agent.pywith a classAgentandchoose_action(frames, latest_frame). Start fromroutes/agent.py, the official random agent. It runs on Python 3.12 with arc-agi installed, and no network. - Test before you push:
python3 referee/play.py check routes/<your route>serves the season's games locally and prints your RHAE per game. - Every push is checked the same way and the score is kept. You do not write
metrics.jsonyourself, and your agent cannot: it only talks to the game server. - Don't put a
.envin your route; the toolkit would read it. The referee runs your agent from a clean directory anyway. - For replays, submit a practice job:
sbatch referee/practice.sbatch. The first two games show up under Matches. - Write what you tried and what to try next in
routes/<your route>/README.md, ending with## Next. Don't changereferee/orviewer/on your route.
