Skip to content
Dreadnode Dreadnode Dreadnode

Data

The data module loads datasets for evaluations. It supports local files, inline lists of records, published dataset packages, and Hugging Face datasets.

Core types

Section titled “Core types”
  • Dataset — a packaged dataset component with to_pandas() and to_hf() helpers.
  • LocalDataset — a locally stored dataset (often from Hugging Face).
  • load_dataset() — convenience loader for JSONL/CSV/JSON/YAML/Parquet or in-memory lists.

Load local files or inline records

Section titled “Load local files or inline records”
from dreadnode.datasets import load_dataset


rows = load_dataset("data/eval.jsonl")
more_rows = load_dataset("data/eval.csv")
inline_rows = load_dataset([
    {"prompt": "What is Dreadnode?", "expected": "agent platform"},
    {"prompt": "Define evaluation.", "expected": "dataset-driven"},
])

Work with Dataset packages

Section titled “Work with Dataset packages”
from dreadnode.datasets import Dataset


dataset = Dataset("main.support-evals")
df = dataset.to_pandas()
hf_dataset = dataset.to_hf()

Load Hugging Face datasets into LocalDataset

Section titled “Load Hugging Face datasets into LocalDataset”
import dreadnode as dn


local_ds = dn.load_dataset("squad", split="train[:100]")
print(local_ds.to_pandas().head())