dwidlee/systemone-lite-general

1---2license: mit3task_categories:4- text-classification5language:6- en7tags:8- system-one9- typed-decisions10- synthetic11size_categories:12- 10K<n<100K13pretty_name: systemone-lite-general14configs:15- config_name: default16  data_files:17  - split: train18    path: data/train-*19  - split: test20    path: data/test-*21  - split: test_hard22    path: data/test_hard-*23  - split: full24    path: data/full-*25---26 27# systemone-lite-general28 29Synthetic typed-decision rows for30[`systemone-lite`](https://github.com/fritzprix/systemone-lite)31(letter-alias `choice` labels for causal LM SFT).32 33Not affiliated with TypeSafe AI / Jev. Labels are rule-based, not human prefs.34 35## Splits36 37| Split | Rows | Notes |38|------|------|------|39| `train` | 32 400 | Stratified mix of 3 gyms |40| `test` | 3 600 | iid held-out by task |41| `test_hard` | 5 400 | layout / paraphrase / option-subset shift |42| `full` | 36 000 | train + iid test |43 44## Gyms45 46- **TicketDungeon**: `ticket.route`, `ticket.needs_human`, `ticket.urgency`47- **ResourceAllocator**: `alloc.fund_next`, `alloc.can_fund_all`, `alloc.pressure`48- **DebateJudge**: `debate.winner`, `debate.enough_evidence`, `debate.confidence`49 50## Downstream numbers (same repo)51 52Fine-tune → [`dwidlee/systemone-lite-0.5b`](https://huggingface.co/dwidlee/systemone-lite-0.5b):53 54| Split | Base 0.5B | SFT |55|---|---:|---:|56| `test` (n=3600) | 0.439 | 0.679 |57| `test_hard` (n=5400) | 0.427 | 0.652 |58 59## Load60 61```python62from datasets import load_dataset63ds = load_dataset("dwidlee/systemone-lite-general")64print(ds)65```66