Datasets:

maxgfr
/

skycmd-zero-shot

license: cc-by-4.0
language:
  - en
  - fr
pretty_name: skycmd zero-shot intent sets
task_categories:
  - text-classification
tags:
  - skycmd
  - typed-decisions
  - system-one
configs:
  - config_name: zs_banking77
    data_files: data/zs_banking77.parquet
  - config_name: zs_massive_en
    data_files: data/zs_massive_en.parquet
  - config_name: zs_massive_fr
    data_files: data/zs_massive_fr.parquet

skycmd zero-shot intent sets

The zero-shot test sets of the s4 and s6 models: MASSIVE intents in English (2,000) and French (1,000) and Banking77 (2,000) as typed choice questions over intents.

Part of skycmd, tiny decision models trained from scratch in Rust that run in the browser. Written by sky-data s4-corpus --only-heldout at commit 6cbcce8c. Licences of every set: DATA_LICENSES.md.

Each row is one typed decision question (docs/SPEC.md):

field	type	meaning
`src`	string	source
`t`	string	`choice`, `score` or `noul`
`ins`	string	the question
`opts`	list of strings	the options, rendered as text
`state`	string	the state the question is about
`target`	list of floats	a probability per option (soft or one-hot)
`split`	string	`train`, `val` or `test`

Source and licence

MASSIVE (CC-BY-4.0) and PolyAI/banking77 (CC-BY-4.0). The conversion is ours.

Files

file	rows
`data/zs_banking77.parquet`	2,000
`data/zs_massive_en.parquet`	2,000
`data/zs_massive_fr.parquet`	1,000

from datasets import load_dataset
ds = load_dataset("maxgfr/skycmd-zero-shot", "zs_banking77")

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.