app.py

import os
import subprocess
from huggingface_hub import hf_hub_download

MODEL_REPO = "gustajunq/OpenFable-4B-GGUF"
MODEL_FILE = "openfable-4b-q4_k_m.gguf"
MODEL_ALIAS = "openfable"

print("Baixando modelo...")

model_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=MODEL_FILE,
)

print("Modelo:", model_path)

cmd = [
    "python",
    "-m",
    "llama_cpp.server",
    "--model",
    model_path,
    "--model_alias",
    MODEL_ALIAS,
    "--host",
    "0.0.0.0",
    "--port",
    "7860",
]

print("Iniciando servidor...")
print(" ".join(cmd))

subprocess.run(cmd)