Qwen2.5-0.5B-Instruct搭建
模型地址
https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct
- 简介
通义千问新一代开源模型Qwen2.5，旗舰模型Qwen2.5-72B性能超越Llama 405B，再登全球开源大模型王座。Qwen2.5全系列涵盖多个尺寸的大语言模型、多模态模型、数学模型和代码模型，每个尺寸都有基础版本、指令跟随版本、量化版本，总计上架100多个模型，刷新业界纪录。 - 环境查看
系统环境
# lsb_release -a
No LSB modules are available.
Distributor ID:Ubuntu
Description:Ubuntu 22.04.5 LTS
Release:22.04
Codename:jammy
# uname -a
Linux AiServer003187 6.8.0-47-generic #47~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Wed Oct 2 16:16:55 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
软件环境
# nvidia-smi
Tue Nov 19 15:36:19 2024
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.120 Driver Version: 550.120 CUDA Version: 12.4 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 4090 Off | 00000000:18:00.0 Off | Off |
| 30% 32C P8 30W / 450W | 18580MiB / 24564MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 1 NVIDIA GeForce RTX 4090 Off | 00000000:3B:00.0 Off | Off |
| 30% 31C P8 35W / 450W | 1366MiB / 24564MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
| 2 NVIDIA GeForce RTX 4090 Off | 00000000:86:00.0 Off | Off |
| 30% 32C P8 19W / 450W | 524MiB / 24564MiB | 0% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 2623 G /usr/lib/xorg/Xorg 4MiB |
| 0 N/A N/A 954745 C /root/anaconda3/bin/python 808MiB |
| 0 N/A N/A 1677511 C /usr/local/bin/python 17736MiB |
| 1 N/A N/A 2623 G /usr/lib/xorg/Xorg 4MiB |
| 1 N/A N/A 28363 C python 492MiB |
| 1 N/A N/A 954745 C /root/anaconda3/bin/python 852MiB |
| 2 N/A N/A 2623 G /usr/lib/xorg/Xorg 4MiB |
| 2 N/A N/A 954745 C /root/anaconda3/bin/python 506MiB |
+-----------------------------------------------------------------------------------------+
# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2024 NVIDIA Corporation
Built on Tue_Feb_27_16:19:38_PST_2024
Cuda compilation tools, release 12.4, V12.4.99
Build cuda_12.4.r12.4/compiler.33961263_0
- 搭建
- 创建conda虚拟环境
# conda create -n qwen python=3.12
- 示例代码
# cat main_test.py
from transformers import AutoModelForCausalLM, AutoTokenizer
# 自动下载模型
#model_name = "Qwen/Qwen2.5-0.5B-Instruct"
local_model_path = "model/Qwen2.5-0.5B-Instruct"
# 如果已经下载好模型可定义模型文件夹
#model_name = "model/Qwen2.5-0.5B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
local_model_path,
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(local_model_path)
prompt = "Give me a short introduction to large language model."
#prompt = "你是谁"
messages = [
{"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
generated_ids = model.generate(
**model_inputs,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(response)
运行输出
运行会下载模型以及依赖
手动下载模型
# pip install modelscope
# modelscope download --model Qwen/Qwen2.5-0.5B-Instruct
- 兼容OpenAI接口
安装软件及依赖
# pip install fastapi
# pip install transformers
# pip install torch
# pip install 'accelerate>=0.26.0'
主程序
# cat main.py
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
from typing import List
# FastAPI 实例
app = FastAPI()
# 定义请求体结构
class Message(BaseModel):
role: str
content: str
class RequestBody(BaseModel):
model: str
messages: List[Message]
max_tokens: int = 100
# 本地模型路径
local_model_path = "model/Qwen2.5-0.5B-Instruct" # 替换为你的本地模型路径
# 加载本地模型和 tokenizer
model = AutoModelForCausalLM.from_pretrained(
local_model_path,
torch_dtype=torch.float16, # 根据实际硬件环境调整
device_map="auto" # 根据设备自动分配
)
tokenizer = AutoTokenizer.from_pretrained(local_model_path)
# 生成文本的 API 路由
@app.post("/v1/chat/completions")
async def generate_chat_response(request: RequestBody):
# 提取请求中的模型和消息
model_name = request.model
messages = request.messages
max_tokens = request.max_tokens
# 构造消息格式（转换为 OpenAI 的格式）
# 使用点语法来访问 Message 对象的属性
combined_message = "\n".join([f"{message.role}: {message.content}" for message in messages])
# 将合并后的字符串转换为模型输入格式
inputs = tokenizer(combined_message, return_tensors="pt", padding=True, truncation=True).to(model.device)
try:
# 生成模型输出
generated_ids = model.generate(
**inputs,
max_new_tokens=max_tokens
)
# 解码输出
response = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
# 格式化响应为 OpenAI 风格
completion_response = {
"id": "some-id", # 你可以根据需要生成唯一 ID
"object": "text_completion",
"created": 1678157176, # 时间戳（可根据实际需求替换）
"model": model_name,
"choices": [
{
"message": {
"role": "assistant",
"content": response
},
"finish_reason": "stop",
"index": 0
}
]
}
return completion_response
except Exception as e:
raise HTTPException(status_code=500, detail=str(e))
# 启动 FastAPI 应用
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
启动
(qwen)root@# python main.py
curl 发送POST请求测试
# curl -X 'POST' 'http://127.0.0.1:8000/v1/chat/completions' -H 'Content-Type: application/json' -d '{
"model": "Qwen/Qwen2.5-0.5B-Instruct",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": "你是谁"}
],
"max_tokens": 100
}'
响应
