|
Download README.md from distributedtensor/posttrain-arena-env: direct link, hf CLI and curl.
- Browser
- Download file 2.5 kB
-
https://huggingface.co/datasets/distributedtensor/posttrain-arena-env/resolve/main/README.md
- Command line
-
hf download hf://datasets/distributedtensor/posttrain-arena-env/README.md
-
curl -L -o README.md https://huggingface.co/datasets/distributedtensor/posttrain-arena-env/resolve/main/README.md
2.5 kB
| license: mit | |
| task_categories: | |
| - reinforcement-learning | |
| # OpenEnv Arena environment: procedural tool-use tasks | |
| One OpenEnv server serving procedurally generated tasks. Each episode is a multi-turn JSON tool conversation (`ls`, `read`, `grep`, `python`, `write`, `submit`, `finish`) in a fresh workspace. Instances are generated at reset from a time-bucketed seed, so the tasks are data inside the image, not files in this dataset. | |
| ## Reward | |
| The terminal reward is `0.6 * (fraction of graded parts correct) + 0.4 * (all parts correct)`, in [0, 1]. Tampering with protected files, or writing nothing, scores 0. The grader runs server-side and the answers are never in the workspace. | |
| ## Provenance | |
| - Source commit: `b593042aac174c32b79409062c80f749e69e8301` (working tree had uncommitted changes in env/ or submit/) | |
| - Image reference at build time: `ghcr.io/distributedstatemachine/arena-env:m4-a` (variant A) | |
| ## Tasks | |
| | task_id | domain | family | level | | |
| |---|---|---|---| | |
| | `f1-industrial` | industrial-physical-systems | late_caveat | 3 | | |
| | `f1-office` | office-white-collar | late_caveat | 3 | | |
| | `f1-security` | cybersecurity | late_caveat | 3 | | |
| | `f1-software` | software-engineering | late_caveat | 3 | | |
| | `f2-industrial` | industrial-physical-systems | false_premise | 5 | | |
| | `f2-media` | media-content-production | false_premise | 4 | | |
| | `f2-office` | office-white-collar | false_premise | 4 | | |
| | `f2-software` | software-engineering | false_premise | 4 | | |
| | `f3-finance` | finance-economics | compute | 4 | | |
| | `f3-industrial` | industrial-physical-systems | compute | 5 | | |
| | `f3-media` | media-content-production | compute | 4 | | |
| | `f3-office` | office-white-collar | compute | 3 | | |
| | `f3-science` | natural-science | compute | 7 | | |
| | `f4-finance` | finance-economics | budget | 4 | | |
| | `f4-industrial` | industrial-physical-systems | budget | 7 | | |
| | `f4-media` | media-content-production | budget | 5 | | |
| | `f4-security` | cybersecurity | budget | 7 | | |
| | `f4-software` | software-engineering | budget | 5 | | |
| | `f5-finance` | finance-economics | artifact | 7 | | |
| | `f5-media` | media-content-production | artifact | 5 | | |
| | `f5-office` | office-white-collar | artifact | 7 | | |
| | `f5-science` | natural-science | artifact | 7 | | |
| | `f5-security` | cybersecurity | artifact | 6 | | |
| | `f7-industrial` | industrial-physical-systems | repair | 4 | | |
| | `f7-media` | media-content-production | repair | 5 | | |
| | `f7-office` | office-white-collar | repair | 4 | | |
| | `f7-software` | software-engineering | repair | 5 | | |