"""Descarga datasets y modelos para la validación externa de DignityAI."""
from datasets import load_dataset
from huggingface_hub import snapshot_download
import os, json

os.makedirs("data", exist_ok=True)
DS = {
    "machismodetect": "antonn-dromundo/MachismoDetect",
    "sinodio_latam": "antonn-dromundo/SinOdio-LATAM-Regional-HateSpeech",
    "piuba_real": "piuba-bigdata/contextualized_hate_speech",
    "hatecheck_es": "Paul/hatecheck-spanish",
    "superset_es": "manueltonneau/spanish-hate-speech-superset",
}
info = {}
for k, rid in DS.items():
    try:
        d = load_dataset(rid)
        for split, part in d.items():
            part.to_pandas().to_parquet(f"data/{k}__{split}.parquet")
        info[k] = {s: {"filas": len(p), "columnas": p.column_names} for s, p in d.items()}
    except Exception as e:
        info[k] = {"error": repr(e)[:300]}
    print(k, json.dumps(info[k], ensure_ascii=False)[:400], flush=True)

for rid in ["piuba-bigdata/beto-contextualized-hate-speech", "antonn-dromundo/SinOdio-BETO-HateSpeech-Detector-v3"]:
    p = snapshot_download(rid, allow_patterns=["*.json", "*.safetensors", "*.txt", "*.md", "tokenizer*", "vocab*", "special*"])
    print(rid, p, os.listdir(p), flush=True)
json.dump(info, open("data/info.json", "w"), ensure_ascii=False, indent=2)