curvedinf/small-qa-1m

---
language:
- en
license: mit
pretty_name: small-qa-autocomplete
task_categories:
- text-generation
task_ids:
- language-modeling
tags:
- synthetic
- autocomplete
- command-line
- unit-conversion
- fact-lookup
size_categories:
- 100K<n<1M
---

# small-qa-autocomplete

Synthetic `prompt -> completion` dataset for training/autocomplete workflows.
Prompts are plain query text designed to resemble short user search/CLI-style requests.

## Dataset Summary

- task type: next-token / completion-style supervision
- source: synthetic prompt permutations with deterministic and model-backed answering
- total rows: 999495

## Splits

- train: 979385
- validation: 10032
- test: 10078

## Schema

Each row includes:

- `id`: stable row id
- `prompt`: user query text
- `completion`: target completion text
- `domain`: domain family
- `intent_id`: prompt intent identifier
- `style_id`: prompt style variant
- `template_id`: generator template identifier
- `metadata`: JSON string with generation metadata

## Recommended Usage

```python
from datasets import load_dataset

# parquet
ds = load_dataset("parquet", data_files={
"train": "train.parquet",
"validation": "validation.parquet",
"test": "test.parquet",
})

# jsonl alternative
# ds = load_dataset("json", data_files={
# "train": "train.jsonl",
# "validation": "validation.jsonl",
# "test": "test.jsonl",
# })
```

## Notes

- This dataset is synthetic and may contain occasional noise.
- It is intended as base pretraining/finetuning material for autocomplete-like behavior.