"""Build chat examples from the Bokmaal Qwen3 Magpie release.

Source: versae/magpie-qwen3-235B-A22B-bokmaal
Revision: e47dc5c0f510425ec0ee064657467803e3477edf
"""

from __future__ import annotations

from collections.abc import Iterator
from pathlib import Path

import langid
from datasets import load_dataset

from dynainst.build_utils import build_dataset

SOURCE = "magpie-qwen3-bokmaal"
HF_DATASET = "versae/magpie-qwen3-235B-A22B-bokmaal"
REVISION = "e47dc5c0f510425ec0ee064657467803e3477edf"
ADDED = "2026-09-03"
CREATED = "2025-05-01, 2025-05-31"
SAVE_DIR = Path(__file__).parent

REASONING_CATEGORIES = {"reasoning", "math", "data-analysis"}
NORWEGIAN_LANGUAGE_LABELS = {"nb", "nn", "no"}

# Restrict the classifier to the languages relevant to this release. The upstream
# dataset contains English prompt/response generations despite its Bokmaal label.
langid.set_languages(["en", *sorted(NORWEGIAN_LANGUAGE_LABELS)])


def task_for(category: str) -> str:
    if category == "coding":
        return "code"
    if category in REASONING_CATEGORIES:
        return "reasoning"
    if category == "editing":
        return "instruction_following"
    return "general"


def convert(raw) -> Iterator[dict]:
    for index, row in enumerate(raw):
        instruction = row["instruction"].strip()
        response = row["response"].strip()
        if langid.classify(instruction)[0] not in NORWEGIAN_LANGUAGE_LABELS:
            continue
        if langid.classify(response)[0] not in NORWEGIAN_LANGUAGE_LABELS:
            continue

        messages = []
        if row["system"].strip():
            messages.append({"role": "system", "content": row["system"].strip()})
        messages.extend(
            [
                {"role": "user", "content": instruction},
                {"role": "assistant", "content": response},
            ]
        )
        yield {
            "id": f"{SOURCE}_{index:05d}",
            "messages": messages,
            "source": SOURCE,
            "added": ADDED,
            "created": CREATED,
            "language": ["nob"],
            "task": task_for(row["category"]),
            "token_count": -1,
        }


def main() -> None:
    raw = load_dataset(HF_DATASET, revision=REVISION, split="train")
    output_path = SAVE_DIR / f"{SOURCE}.parquet"
    dataset = build_dataset(convert(raw), output_path=output_path)
    print(f"Wrote {len(dataset):,} samples to {output_path}")


if __name__ == "__main__":
    main()