---
dataset_info:
  features:
  - name: award_id
    dtype: string
  - name: technical_abstract
    dtype: string
  - name: non_technical_abstract
    dtype: string
  - name: verifiable_claims
    sequence: string
  - name: investigation_proposals
    sequence: string
  - name: award_year
    dtype: float64
  - name: division
    dtype: string
  - name: directorate
    dtype: string
  - name: title
    dtype: string
  splits:
  - name: train
    num_examples: 14682
  - name: validation
    num_examples: 500
  - name: test
    num_examples: 4000
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*.parquet
  - split: validation
    path: data/validation-*.parquet
  - split: test
    path: data/test-*.parquet
---

# NSF-SciFy 20K

Filtered 20K NSF award dataset with technical abstracts, non-technical abstracts,
verifiable claims, and investigation proposals.

## Load

```python
from datasets import load_dataset

ds = load_dataset("darpa-scify/nsf-scify-20k")
```

## Splits

- `train`: 14,682 examples
- `validation`: 500 examples
- `test`: 4,000 examples