Upload a reasoning dataset¶

Reasoning datasets power Advanced Training experiments that learn from prompt–response examples.

Prerequisites¶

An Advanced Training project.
A CSV or supported file layout for reasoning data (see Reasoning datasets).

2. New dataset¶

Click New dataset and choose the reasoning dataset type when prompted.

Create dataset modal with reasoning type — Creating a new reasoning dataset.¶

3. Upload files¶

Upload local files or choose a reasoning dataset from Hugging Face. Follow on-screen column hints for reasoning fields.

Reasoning dataset file upload — Uploading reasoning dataset files.¶

4. Map columns¶

Map your file columns to Question and Answer using the column label controls.

Reasoning dataset column mapping step — Selecting question and answer columns.¶

5. Preview¶

Review a sample table after column mapping.

Dataset preview table — Preview of uploaded reasoning rows.¶

6. Saved dataset¶

Confirm save. The dataset appears in the list for use in the experiment wizard.

Datasets list with new reasoning dataset — The dataset listed after save.¶

From the terminal¶

Create the dataset.

Install the CLI and sign in (Authentication):

pip install "agilerl[arena]"
arena login
# or: export ARENA_API_KEY=arena_pat_…

arena datasets create math-word-problems --category reasoning --column-mapping '{"question":"question","answer":"answer"}' --file ./reasoning.csv

category must be reasoning. Mapping keys are Arena names (question, answer); values are column names in your file.

Python¶

from agilerl.arena import ArenaClient

with ArenaClient() as client:
    client.create_dataset(
        name="math-word-problems",
        category="reasoning",
        file="./reasoning.csv",
        column_mapping={"question": "question", "answer": "answer"},
    )