大模型部署实践2

第27-28 学时：大模型任务实践

• 课中实施：

  • 使用python部署deepseek R1 1.5B

  • 使用deepseek帮我们写代码。
    

• 软件环境

• python-3.11.7-amd64

• PyCharm IDE

• 下载地址：
  python-3.11.7
  PyCharm IDE
  

• 实践步骤

• 1.下载模型代码

import os
import requests

base_url = "http://192.168.189.3:8015/DeepSeek-R1-Distill-Qwen-1.5B/"
files = [
    "config.json",
    "generation_config.json",
    "tokenizer_config.json",
    "tokenizer.json",
    "model.safetensors"
]

save_dir = r"D:\aimodel\DeepSeek-R1-Distill-Qwen-1.5B"
os.makedirs(save_dir, exist_ok=True)

for file in files:
    url = base_url + file
    save_path = os.path.join(save_dir, file)
    print(f"⬇️ 正在下载 {file} ...")
    r = requests.get(url, stream=True)
    with open(save_path, "wb") as f:
        for chunk in r.iter_content(chunk_size=8192):
            if chunk:
                f.write(chunk)
print("✅ 所有模型文件下载完成！")

• 2.运行模型测试代码

• 
  

  javascripttypescripthtmlcssshellpythongolangjavacc++c#phprubyswiftkotlinscalarustdartelixirhaskellluaperlrsql

  from transformers import AutoTokenizer, AutoModelForCausalLM
  import torch
  
  # 模型路径（你本地下载的）
  model_path = "D:/aimodel/DeepSeek-R1-Distill-Qwen-1.5B"
  
  # 加载 tokenizer 和 model
  print("📦 正在加载模型和分词器...")
  tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
  model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)
  model.eval()
  
  # 如果有GPU，放上去
  device = "cuda" if torch.cuda.is_available() else "cpu"
  model.to(device)
  print(f"🚀 模型已加载，使用设备：{device}")
  
  # 初始化对话历史
  chat_history = []
  
  def build_prompt(history, new_question):
      full_prompt = ""
      for i, (q, a) in enumerate(history):
          full_prompt += f"[Round {i+1}]\n问：{q}\n答：{a}\n"
      full_prompt += f"[Round {len(history)+1}]\n问：{new_question}\n答："
      return full_prompt
  
  print("\n🤖 欢迎使用 DeepSeek 模型！输入你的问题吧，输入 `exit` 退出。\n")
  
  while True:
      query = input("🧑 你：")
      if query.strip().lower() in ["exit", "quit"]:
          print("👋 再见！")
          break
  
      prompt = build_prompt(chat_history, query)
      inputs = tokenizer(prompt, return_tensors="pt").to(device)
  
      with torch.no_grad():
          outputs = model.generate(
              **inputs,
              max_new_tokens=200,
              do_sample=True,
              temperature=0.7,
              top_p=0.9,
              repetition_penalty=1.1,
              eos_token_id=tokenizer.eos_token_id,
          )
  
      response = tokenizer.decode(outputs[0], skip_special_tokens=True)
      answer = response[len(prompt):].strip()
  
      print(f"🤖 DeepSeek：{answer}\n")
      chat_history.append((query, answer))
  

• 3. 常见问题是下载不了软件包

• 解决办法：安装时指定镜像仓库源

• 清华大学pypi软件仓库： https://pypi.tuna.tsinghua.edu.cn/simple

• 阿里pypi软件仓库： https://mirrors.aliyun.com/pypi/simple/

• 腾讯pypi软件仓库：https://mirrors.cloud.tencent.com/pypi/simple/

• 使用方法：
  

  javascripttypescripthtmlcssshellpythongolangjavacc++c#phprubyswiftkotlinscalarustdartelixirhaskellluaperlrsql

  pip install requests -i https://mirrors.cloud.tencent.com/pypi/simple