THUDM / CodeGeeX2

CodeGeeX2: A More Powerful Multilingual Code Generation Model
https://codegeex.cn
Apache License 2.0
7.55k stars 532 forks source link
code code-generation pretrained-models tool

🏠 主页|🛠 插件 VS Code, Jetbrains|🤗 模型下载|📄 论文|👋 加入微信开发者交流群

Read this in English
日本語で読む
Lire en Français

CodeGeeX2: 更强大的多语言代码生成模型

CodeGeeX2 是多语言代码生成模型 CodeGeeX (KDD’23) 的第二代模型。不同于一代 CodeGeeX(完全在国产华为昇腾芯片平台训练) ,CodeGeeX2 是基于 ChatGLM2 架构加入代码预训练实现,得益于 ChatGLM2 的更优性能,CodeGeeX2 在多项指标上取得性能提升(+107% > CodeGeeX;仅60亿参数即超过150亿参数的 StarCoder-15B 近10%),更多特性包括:

使用教程

AI编程助手

我们开发了支持 VS Code、 IntelliJ IDEA、PyCharm、GoLand、WebStorm、Android Studio 等IDE的 CodeGeeX 插件。在插件中,可以更直接地体验到 CodeGeeX2 模型在代码生成与补全、添加注释、代码翻译及技术问答方面的能力为开发效率带来的提升。欢迎在IDE中下载 CodeGeeX 插件获得更加全面的AI编程体验,详情见CodeGeeX主页

快速开始

使用transformers快速调用CodeGeeX2-6B

from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("THUDM/codegeex2-6b", trust_remote_code=True)
model = AutoModel.from_pretrained("THUDM/codegeex2-6b", trust_remote_code=True, device='cuda')
model = model.eval()

# remember adding a language tag for better performance
prompt = "# language: Python\n# write a bubble sort function\n"
inputs = tokenizer.encode(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_length=256, top_k=1)
response = tokenizer.decode(outputs[0])

>>> print(response)
# language: Python
# write a bubble sort function

def bubble_sort(list):
    for i in range(len(list) - 1):
        for j in range(len(list) - 1):
            if list[j] > list[j + 1]:
                list[j], list[j + 1] = list[j + 1], list[j]
    return list

print(bubble_sort([5, 2, 1, 8, 4]))

启动 Gradio DEMO:

python ./demo/run_demo.py

usage: run_demo.py [-h] [--model-path MODEL_PATH] [--example-path EXAMPLE_PATH] [--quantize QUANTIZE]
                   [--chatglm-cpp] [--fastllm] [--n-gpus N_GPUS] [--gpu GPU] [--cpu] [--auth] [--username yourname]
                   [--password yourpassword]
                   [--port PORT] [--listen ADDRESS]

# 若要启用身份验证,请先启用--auth,然后定义--username与--password,如:
python run_demo.py --auth --username user --password password  # 若要监听所有地址请指定 --listen 0.0.0.0

支持使用 ChatGLM.cpp 量化推理加速:

python ./demo/run_demo.py --quantize 4 --chatglm-cpp

启动FAST API:

python ./demo/fastapicpu.py
usage: fastapicpu.py [-h] [--model-path MODEL_PATH] [--listen ADDRESS] [--port PORT] [--workders NUM] [--cpu] [--half] [--quantize QUANTIZE] [--chatglm-cpp]
# --cpu启用cpu --half启用.half()

支持使用 ChatGLM.cpp 量化推理加速,同样添加 --quantize 4 --chatglm-cpp 参数即可。

API使用示例

curl -X POST "http://127.0.0.1:7860" \
    -H 'Content-Type: application/json' \
    -d '{"lang": "Python", "prompt": "# Write a quick sort function"}'

❗️请注意:

代码能力评测

CodeGeeX2 作为一个多语言代码生成基座模型,代码能力较上一代大幅提升,以下是在 HumanEval,HumanEval-X, DS1000 基准上的评测结果(评价指标 Pass@k 定义与论文中一致):

HumanEval (Pass@1,10,100)

Model Pass@1 Pass@10 Pass@100
CodeGen-16B-multi 19.2 34.6 55.2
CodeGeeX-13B 22.9 39.6 60.9
Codex-12B 28.8 46.8 72.3
CodeT5Plus-16B-mono 30.9 51.6 76.7
Code-Cushman-001 33.5 54.3 77.4
LLaMA-65B 23.7 - 79.3
LLaMA2-70B 29.9 - -
CodeGen2.5-7B-mono 33.4 58.4 82.7
StarCoder-15B 33.2 61.0 84.7
CodeGeeX2-6B 35.9 62.6 88.3

Pass@1 使用 n=20, t=0.2, top_p=0.95Pass@10,Pass@100 使用 n=200, t=0.8, top_p=0.95

HumanEval-X (Pass@1)

Model Python C++ Java JavaScript Go Rust Overall
CodeGen-16B-multi 19.2 18.1 15.0 18.4 13.0 1.8 14.2
CodeGeeX-13B 22.9 17.1 20.0 17.6 14.4 4.3 16.0
Replit-code-v1-3B 22.0 20.1 20.1 20.1 12.2 8.6 17.2
CodeGen2.5-7B-multi 30.6 24.3 29.0 27.5 18.9 20.1 25.1
StarCoder-15B 35.5 28.2 31.5 33.2 21.3 17.8 27.9
CodeGeeX2-6B 35.9 29.3 30.8 32.2 22.5 18.1 28.1

Pass@1 使用 n=20, t=0.2, top_p=0.95

以上结果可使用脚本scripts/run_humanevalx.sh复现。环境配置和说明参见评测环境

DS1000 (Pass@1)

Model Matplotlib Numpy Pandas Pytorch SciPy Scikit-learn TensorFlow Overall
# Samples 155 220 291 68 106 115 45 1000
CodeGen-16B-Mono 31.7 10.9 3.4 7.0 9.0 10.8 15.2 11.7
code-cushman-001 40.7 21.8 7.9 12.4 11.3 18.0 12.2 18.1
Codex-001 41.8 26.6 9.4 9.7 15.0 18.5 17.2 20.2
CodeGeeX2-6B 40.5 25.5 14.5 17.3 19.3 24.0 23.0 23.1
StarCoder-15B 51.7 29.7 11.4 21.4 20.2 29.5 24.5 26.0
Codex-002 57.0 43.1 26.5 41.8 31.8 44.8 39.3 39.2

Pass@1 使用 n=40, t=0.2, top_p=0.5

以上结果可使用DS1000评测代码复现。

量化推理性能

CodeGeeX2 与上一代相比,对部署更加友好。得益于使用 Multi-Query Attention 和 Flash Attention,推理速度更快,且量化后仅需6GB显存即可运行:

量化

Model FP16/BF16 INT8 INT4
CodeGeeX-13B 26.9 GB 14.7 GB -
CodeGeeX2-6B 13.1 GB 8.2 GB 5.5 GB

基于 PyTorch 2.0 测试,利用torch.nn.functional.scaled_dot_product_attention实现高效的 Attention 计算。

推理

Model 推理速度 (字符/秒)
CodeGeeX-13B 32
CodeGeeX2-6B 94

batch_size=1, max_length=2048,均使用加速框架,测试硬件为GeForce RTX-3090

协议

本仓库的代码依照 Apache-2.0 协议开源,模型的权重的使用则需要遵循 Model License。CodeGeeX2-6B 权重对学术研究完全开放,填写登记表申请商业使用。

引用

如果觉得我们的工作有帮助,欢迎引用以下论文:

@inproceedings{zheng2023codegeex,
      title={CodeGeeX: A Pre-Trained Model for Code Generation with Multilingual Evaluations on HumanEval-X},
      author={Qinkai Zheng and Xiao Xia and Xu Zou and Yuxiao Dong and Shan Wang and Yufei Xue and Zihan Wang and Lei Shen and Andi Wang and Yang Li and Teng Su and Zhilin Yang and Jie Tang},
      booktitle={KDD},
      year={2023}
}