"""Download and verify unchanged base files for redistribution; never train or upload."""
import hashlib
import json
import shutil
from pathlib import Path
from huggingface_hub import HfApi, hf_hub_download

ROOT = Path(__file__).resolve().parents[1]
MODEL = "LiquidAI/LFM2.5-350M"
REVISION = "9e6c6ccf47cd318696e137d381a7ded8fe4df09f"
FILES = ["model.safetensors", "config.json", "generation_config.json", "tokenizer.json", "tokenizer_config.json", "chat_template.jinja", "LICENSE"]


def sha256(path):
    with path.open("rb") as source:
        return hashlib.file_digest(source, "sha256").hexdigest()


def main():
    info = HfApi().model_info(MODEL, revision=REVISION, files_metadata=True)
    metadata = {f.rfilename: f for f in info.siblings}
    code_license = ROOT / "LICENSE-CODE"
    if not code_license.exists():
        assert (ROOT / "LICENSE").read_text().startswith("MIT License")
        shutil.copy2(ROOT / "LICENSE", code_license)
    manifest = {"source_repository": MODEL, "source_revision": REVISION,
                "weights_modified": False, "training_performed": False, "files": {}}
    for name in FILES:
        source = Path(hf_hub_download(MODEL, name, revision=REVISION))
        digest = sha256(source)
        remote = metadata[name]
        if remote.lfs:
            assert digest == remote.lfs.sha256, name
        else:
            content = source.read_bytes()
            blob = hashlib.sha1(f"blob {len(content)}\0".encode() + content).hexdigest()
            assert blob == remote.blob_id, name
        destination = ROOT / name
        shutil.copy2(source, destination)
        assert sha256(destination) == digest, name
        manifest["files"][name] = {"sha256": digest, "size_bytes": destination.stat().st_size}
        print(f"Verified unchanged: {name}", flush=True)
    (ROOT / "BASE_MODEL_MANIFEST.json").write_text(json.dumps(manifest, indent=2) + "\n")

if __name__ == "__main__":
    main()