# Portable AI Runner - one-time setup v4 (resumable + sha256-verified GGUF fetch)
# Online ONCE:  pip install requests
#   python setup.py                -> Qwen3-0.6B GGUF (639 MB) + wllama engine assets
#   python setup.py --model ornith -> also Ornith-1.5-9B GGUF (5.8 GB, server-tier file)
# Offline forever after. GGUF only - no banned engines (owner directive 2026-09-18).
import hashlib
import io
import json
import os
import sys
import tarfile

import requests

WLLAMA_VERSION = "3.6.1"
GGUF_DIR = os.path.join("models", "gguf")

# sha256 = Hub LFS oids, verified 2026-09-18 (models/weights-manifest.json)
PINNED = {
    "Qwen3-0.6B-Q8_0.gguf": (
        "Qwen/Qwen3-0.6B-GGUF", 639446688,
        "9465e63a22add5354d9bb4b99e90117043c7124007664907259bd16d043bb031",
    ),
    "Ornith-1.5-9B-Q4_K_M.gguf": (
        "ornith-ai/Ornith-1.5-9B-GGUF", 5780090816,
        "70c112196e0b7023803c9762752e46d29e612a92c83f995bc3ba1ceb07e8fab6",
    ),
}


def vendor_wllama():
    os.makedirs(os.path.join("vendor", "wllama", "esm"), exist_ok=True)
    print("downloading wllama engine assets (llama.cpp WASM) ...")
    tgz = requests.get(
        "https://registry.npmjs.org/@wllama/wllama/-/wllama-" + WLLAMA_VERSION + ".tgz",
        timeout=120).content
    with tarfile.open(fileobj=io.BytesIO(tgz), mode="r:gz") as t:
        for member in t.getmembers():
            if member.name.startswith("package/esm/"):
                dest = os.path.join("vendor/wllama/esm",
                                    os.path.relpath(member.name, "package/esm"))
                if member.isdir():
                    os.makedirs(dest, exist_ok=True)
                    continue
                os.makedirs(os.path.dirname(dest), exist_ok=True)
                with open(dest, "wb") as f:
                    f.write(t.extractfile(member).read())
    wasm = os.path.join("vendor", "wllama", "esm", "wasm", "wllama.wasm")
    assert os.path.isfile(wasm) and os.path.getsize(wasm) > 1_000_000, \
        "wllama.wasm missing after vendoring"
    print("  ok: vendored, esm/wasm/wllama.wasm present")


def fetch_gguf(repo, fname, size, sha_expected):
    url = "https://huggingface.co/" + repo + "/resolve/main/" + fname
    dest = os.path.join(GGUF_DIR, fname)
    pos = os.path.getsize(dest) if os.path.exists(dest) else 0
    if pos == size:
        print("  already complete:", dest)
    else:
        if pos:
            print("  resuming from %.2f GB ..." % (pos / 1e9))
        headers = {"Range": "bytes=%d-" % pos} if pos else {}
        r = requests.get(url, stream=True, timeout=600, headers=headers)
        if pos and r.status_code != 206:
            print("  server does not support resume - restarting")
            pos, r = 0, requests.get(url, stream=True, timeout=600)
        r.raise_for_status()
        with open(dest, "ab" if pos else "wb") as f:
            for chunk in r.iter_content(1 << 22):
                f.write(chunk)
    print("  verifying sha256 ...")
    h = hashlib.sha256()
    with open(dest, "rb") as f:
        for chunk in iter(lambda: f.read(1 << 22), b""):
            h.update(chunk)
    if h.hexdigest() != sha_expected:
        os.remove(dest)
        raise SystemExit("SHA256 MISMATCH for " + fname +
                         " - corrupt/partial download removed. Re-run setup.py.")
    print("  ok: " + fname + " sha256 verified")


def main():
    args = [a.lower() for a in sys.argv[1:]]
    want_ornith = any("ornith" in a for a in args)
    os.makedirs(GGUF_DIR, exist_ok=True)
    vendor_wllama()
    picks = []
    if want_ornith:
        picks.append("Ornith-1.5-9B-Q4_K_M.gguf")
    picks.append("Qwen3-0.6B-Q8_0.gguf")
    entries = []
    for fname in picks:
        repo, size, sha = PINNED[fname]
        print("downloading " + repo + "/" + fname + " (%.2f GB) ..." % (size / 1e9))
        fetch_gguf(repo, fname, size, sha)
        entries.append({"name": fname.replace(".gguf", ""),
                        "url": "./models/gguf/" + fname,
                        "size_gb": round(size / 1e9, 2), "sha256": sha})
    manifest = {"engine": "wllama (llama.cpp WASM)", "models": entries}
    with open(os.path.join("models", "models.json"), "w") as f:
        json.dump(manifest, f, indent=2)
    print("done. Launch serve.py and open http://localhost:8080 - fully offline from here on.")


if __name__ == "__main__":
    main()