README.md

metadata

license: apache-2.0
language:
  - zh
  - en
library_name: mlx
base_model:
  - Qwen/Qwen2-7B-Instruct
tags:
  - macbook

BiliBot

b友风格聊天机器人

• 基础模型: Qwen2-7B
• 数据来源: https://github.com/linyiLYi/bilibot/tree/main/data
• 量化: 4bit
• 推荐配置: 16G内存及以上的M系芯片Macbook

由于是MLX格式模型，首先需要安装 mlx-lm 包

pip install mlx-lm

下面是一个示例，用户可随意提问

import time
from mlx_lm import load, generate

model, tokenizer = load('Kadins/BiliBot-7B-Q', tokenizer_config={"eos_token": "<|im_end|>"})

# Template content
template = """
<|im_start|>system
You are a helpful assistant<|im_end|>
<|im_start|>user
你是一位B站老用户，请你对以下问题给出简短、机智的回答：
{usr_msg}<|im_end|>
<|im_start|>assistant
"""

while True:
    usr_msg = input("用户: ")  # Get user message from terminal
    if usr_msg.lower() == 'quit()':  # Allows the user to exit the loop
        break

    prompt = template.replace("{usr_msg}", usr_msg)

    time_ckpt = time.time()
    response = generate(
        model,
        tokenizer,
        prompt=prompt,
        temp=0.3,
        max_tokens=500,
        verbose=False
    )

    print("%s: %s (Time %d ms)\n" % ("回答", response, (time.time() - time_ckpt) * 1000))