--- license: apache-2.0 language: - en - fr pretty_name: skycmd hand-written evaluation sets task_categories: - text-classification tags: - skycmd configs: - config_name: demo-cases data_files: data/demo-cases.parquet --- # skycmd hand-written evaluation sets The evaluation material skycmd wrote by hand: the cases of the browser demos, the questions and class definitions of the s5 zero-shot sets, and the paraphrases of the typed-decisions test definitions. Part of [skycmd](https://github.com/maxgfr/skycmd), tiny decision models trained from scratch in Rust that run in the browser. Written by `written by hand, frozen by docs/s5-prereg.md` at commit [`8cc8e276`](https://github.com/maxgfr/skycmd/commit/8cc8e276adabbe1cc37998fb70fdd5c947ffda81). Licences of every set: [DATA_LICENSES.md](https://github.com/maxgfr/skycmd/blob/main/DATA_LICENSES.md). `demo-cases` has the typed-decision schema with `opts_kv` (pairs of option id and definition) instead of `opts`. Two files are frozen by `docs/s5-prereg.md` (their sha256 is recorded there) and shipped as they are: - `files/defs.json`: the hand-written questions and class definitions of the s5 zero-shot sets; - `files/paraphrases.json`: hand-written paraphrases of the option definitions of the typed-decisions test split, with no template in common with the training paraphrases. ## Sources and licences Each config is one source and keeps **its source's licence** (share-alike sources stay share-alike). Attribution goes to the upstream authors; the conversion is ours. | config | upstream | licence | how it is used | rows | |---|---|---|---|---:| | `demo-cases` | ours, hand-written (`eval/s5/demo-cases.jsonl`) | apache-2.0 | the 56 cases of the browser demos (30 en, 26 fr): tools, guard, triage, score, playground; excluded from training | 56 | ## Files | file | rows | |---|---:| | `data/demo-cases.parquet` | 56 | | `files/defs.json` | — | | `files/paraphrases.json` | — | ```python from datasets import load_dataset ds = load_dataset("maxgfr/skycmd-evals", "demo-cases") ``` skycmd wrote no row with a language model: the targets come from the upstream labels (some upstream sets carry their own teacher soft labels, see their cards), the games' oracles or our generators.