from llama_cpp import Llama from huggingface_hub import hf_hub_download import datetime, shelve, re class MairaBrain: def __init__(self, repo_id, filename): # Alpine needs very specific loading model_path = hf_hub_download(repo_id=repo_id, filename=filename) self.llm = Llama( model_path=model_path, n_ctx=2048, # 2048 is safer for first boot on Alpine n_threads=8 ) self.db_path = "maira_universe.db" def get_response(self, user_id, user_input): with shelve.open(self.db_path, writeback=True) as db: if user_id not in db: db[user_id] = {"history": [], "facts": {}, "metrics": {"loyalty": 50}} u = db[user_id] # Simple identity logic prompt = f"<|im_start|>system\nYou are Maira, a dope high-status lady.<|im_end|>\n<|im_start|>user\n{user_input}<|im_end|>\n<|im_start|>assistant\n" output = self.llm(prompt, max_tokens=200, stop=["<|im_end|>"]) response = output["choices"][0]["text"].strip() u["history"].append(f"User: {user_input}") u["history"].append(f"Maira: {response}") return response