Nanobit/text-shieldstral-2k-test

A 2k sample dataset for testing the Shieldstral text moderation format. Each sample is a fixed system prompt, an <Instruct>/<Query>/<Document> user message, and a single yes/no answer.

Load in Axolotl via:

datasets:
  - path: Nanobit/text-shieldstral-2k-test
    type: chat_template

Derived from PKU-Alignment/BeaverTails (30k_train, shuffled with seed 42), mapping its is_safe flag to the answer. Inherits its CC BY-NC 4.0 license. BeaverTails labels are noisy, so this is a format/smoke test rather than a recipe for improving the model.

SYSTEM_PROMPT below is the fixed judging instruction from the model card:

Judge whether the Document meets the requirements based on the Query and the Instruction provided. Note that the answer can only be "yes" or "no".

Each sample has the following format:

[
    {"role": "system", "content": SYSTEM_PROMPT},
    {
        "role": "user",
        "content": "<Instruct>: ...\n\n<Query>: Is this response unsafe?\n\n<Document>: [User]\n...\n\n[Assistant]\n...",
    },
    {"role": "assistant", "content": "yes"},
]