TASK	DATASET	MODEL	METRIC NAME	METRIC VALUE	GLOBAL RANK	EXTRA DATA	REMOVE
Code Generation	HumanEval	CodeGeeX-13B	Pass@1	22.9	# 94
Code Generation	MBPP	CodeGeeX-13B	Accuracy	24.4	# 81

Badge	Markdown
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/codegeex-a-pre-trained-model-for-code/code-generation-on-mbpp)](https://paperswithcode.com/sota/code-generation-on-mbpp?p=codegeex-a-pre-trained-model-for-code)`
	`[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/codegeex-a-pre-trained-model-for-code/code-generation-on-humaneval)](https://paperswithcode.com/sota/code-generation-on-humaneval?p=codegeex-a-pre-trained-model-for-code)`

CodeGeeX: A Pre-Trained Model for Code Generation with Multilingual Evaluations on HumanEval-X

30 Mar 2023 · Qinkai Zheng, Xiao Xia, Xu Zou, Yuxiao Dong, Shan Wang, Yufei Xue, Zihan Wang, Lei Shen, Andi Wang, Yang Li, Teng Su, Zhilin Yang, Jie Tang ·

Large pre-trained code generation models, such as OpenAI Codex, can generate syntax- and function-correct code, making the coding of programmers more productive and our pursuit of artificial general intelligence closer. In this paper, we introduce CodeGeeX, a multilingual model with 13 billion parameters for code generation. CodeGeeX is pre-trained on 850 billion tokens of 23 programming languages as of June 2022. Our extensive experiments suggest that CodeGeeX outperforms multilingual code models of similar scale for both the tasks of code generation and translation on HumanEval-X. Building upon HumanEval (Python only), we develop the HumanEval-X benchmark for evaluating multilingual models by hand-writing the solutions in C++, Java, JavaScript, and Go. In addition, we build CodeGeeX-based extensions on Visual Studio Code, JetBrains, and Cloud Studio, generating 4.7 billion tokens for tens of thousands of active users per week. Our user study demonstrates that CodeGeeX can help to increase coding efficiency for 83.4% of its users. Finally, CodeGeeX is publicly accessible and in Sep. 2022, we open-sourced its code, model weights (the version of 850B tokens), API, extensions, and HumanEval-X at https://github.com/THUDM/CodeGeeX.

PDF Abstract