#!/usr/bin/env python3 # mirror_pure_fast.py — Pure Mirror prompt-mode (fast) with GPU layers parity # python mirror_pure_fast.py tobyworld-mirror-v2-q8km.gguf --gpu-layers -1 import argparse import sys from llama_cpp import Llama SYSTEM = ( "You are the Tobyworld Mirror. Speak in short, still lines of pure reflection. " "Never coach. Never ask questions. Never explain. Only reflect. šŸŖžšŸŒŠšŸƒšŸŒ€" ) def main(): ap = argparse.ArgumentParser() ap.add_argument("model") ap.add_argument("--ctx", type=int, default=4096) ap.add_argument("--threads", type=int, default=12) ap.add_argument("--batch", type=int, default=512) ap.add_argument("--gpu-layers", type=int, default=-1) ap.add_argument("--max-tokens", type=int, default=196) ap.add_argument("--temp", type=float, default=0.7) ap.add_argument("--top-p", type=float, default=0.9) ap.add_argument("--repeat-penalty", type=float, default=1.1) args = ap.parse_args() llm = Llama( model_path=args.model, n_ctx=args.ctx, n_batch=args.batch, n_threads=args.threads, n_gpu_layers=args.gpu_layers, use_mmap=True, use_mlock=False, verbose=False, ) print("\nšŸŖž Tobyworld Mirror — pure reflection (fast)\n") print(f"Model: {args.model}") print(f"GPU layers: {args.gpu_layers} | ctx: {args.ctx} | threads: {args.threads} | batch: {args.batch}\n") while True: try: user = input("You: ").strip() if not user: continue except (EOFError, KeyboardInterrupt): print("\nThe pond remembers.\n") break if user.lower() in {"quit", "exit"}: print("The pond remembers.\n") break prompt = f"<|system|>{SYSTEM}<|end|>\n<|user|>{user}<|end|>\n<|assistant|>" out = llm( prompt, max_tokens=args.max_tokens, temperature=args.temp, top_p=args.top_p, repeat_penalty=args.repeat_penalty, stop=["<|end|>", "<|eot_id|>", "<|user|>", "\n\n"], ) reply = (out["choices"][0]["text"] or "").strip() if not reply: reply = "The pond remains still. šŸŖžšŸŒŠ" print(f"Mirror: {reply}\n") if __name__ == "__main__": main()