真棒代码LLM

🧵 Table of Contents

 

• 🧵 Table of Contents
• 🚀 Leaderboard
• 💡 Evaluation Toolkit:
• 📚 Paper
  • ▶️ Pre-Training
  • ▶️ Instruction Tuning
  • ▶️ Alignment with Feedback
  • ▶️ Prompting
  • ▶️ Evaluation & Benchmark
  • ▶️ Using LLMs while coding
• 🙌 Contributors
• Cite as
• Acknowledgement
• Star History

🚀 Leaderboard

 

Central Leaderboard (Sort by HumanEval Pass@1)

Model	Params	HumanEval	MBPP	HF	Source
GPT-4 + Reflexion	?	91.0	77.1		paper
GPT-4 (latest)	?	84.1	80.0		github
DeepSeek-Coder-Instruct	33B	79.3	70.0	ckpt	github
DeepSeek-Coder-Instruct	7B	78.6	65.4	ckpt	github
GPT-3.5-Turbo (latest)	?	76.2	70.8		github
Code-Llama	34B	62.2	61.2		paper
Pangu-Coder2	15B	61.6			paper
WizardCoder-15B	15B	57.3	51.8	ckpt	paper
Code-Davinci-002	?	47.0			paper
StarCoder-15B (Prompted)	15B	40.8	49.5	ckpt	paper
PaLM 2-S	?	37.6	50.0		paper
PaLM-Coder-540B	540B	36.0	47.0		paper
InstructCodeT5+	16B	35.0			paper
StarCoder-15B	15B	33.6	52.7	ckpt	paper
Code-Cushman-001	?	33.5	45.9		paper
CodeT5+	16B	30.9			paper
LLaMA2-70B	70B	29.9		ckpt	paper
CodeGen-16B-Mono	16B	29.3	35.3		paper
PaLM-540B	540B	26.2	36.8		paper
LLaMA-65B	65B	23.7	37.7		paper
CodeGeeX	13B	22.9	24.4		paper
LLaMA-33B	33B	21.7	30.2		paper
CodeGen-16B-Multi	16B	18.3	20.9		paper
AlphaCode	1.1B	17.1			paper

Leaderboard	Access
Big Code Models Leaderboard	[Source]
BIRD	[Source]
CanAiCode Leaderboard	[Source]
Coding LLMs Leaderboard	[Source]
CRUXEval Leaderboard	[Source]
EvalPlus	[Source]
HumanEval.jl	[Source]
InfiCoder-Eval	[Source]
InterCode	[Source]
Program Synthesis Models Leaderboard	[Source]
Spider	[Source]

💡 Evaluation Toolkit:

 

• bigcode-evaluation-harness: A framework for the evaluation of autoregressive code generation language models.
• code-eval: A framework for the evaluation of autoregressive code generation language models on HumanEval.

📚 Paper

 

▶️ Pre-Training

 

1. Evaluating Large Language Models Trained on Code Preprint

   [Paper] Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto. et al. 2021.07

2. CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis ICLR23

   [Paper] Erik Nijkamp, Bo Pang, Hiroaki Hayashi, Lifu Tu, Huan Wang, Yingbo Zhou, Silvio Savarese, Caiming Xiong. 2022.03

3. ERNIE-Code: Beyond English-Centric Cross-lingual Pretraining for Programming Languages ACL23 (Findings)

   [Paper][Repo] Yekun Chai, Shuohuan Wang, Chao Pang, Yu Sun, Hao Tian, and Hua Wu. 2022.12

4. SantaCoder: don't reach for the stars! Preprint

   [Paper] Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff. et al. 2023.01

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

   [Paper] 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. 2023.03

6. CodeGen2: Lessons for Training LLMs on Programming and Natural Languages ICLR23

   [Paper] Erik Nijkamp, Hiroaki Hayashi, Caiming Xiong, Silvio Savarese, Yingbo Zhou. 2023.05

7. StarCoder: may the source be with you! Preprint

   [Paper] Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou. et al. 2023.05

8. CodeT5+: Open Code Large Language Models for Code Understanding and Generation Preprint

   [Paper] Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi D.Q. Bui, Junnan Li, Steven C.H. Hoi. 2023.05

9. Textbooks Are All You Need Preprint

   [Paper] Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio César Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi. et al. 2023.06

10. Code Llama: Open Foundation Models for Code Preprint

    [Paper] Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat. et al. 2023.08

11. DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence Preprint

    [Paper] Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen et al. 2024.01

12. StarCoder 2 and The Stack v2: The Next Generation Preprint

    [Paper] Anton Lozhkov, Raymond Li, Loubna Ben Allal, Federico Cassano, Joel Lamy-Poirier, Nouamane Tazi, Ao Tang et al. 2024.02

▶️ Instruction Tuning

 

1. Code Alpaca: An Instruction-following LLaMA Model trained on code generation instructions

   [Repo] Sahil Chaudhary. 2023

2. WizardCoder: Empowering Code Large Language Models with Evol-Instruct Preprint

   [Paper] Ziyang Luo, Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, Daxin Jiang. 2023.07

3. OctoPack: Instruction Tuning Code Large Language Models Preprint

   [Paper][Repo] Niklas Muennighoff, Qian Liu, Armel Zebaze, Qinkai Zheng, Binyuan Hui, Terry Yue Zhuo, Swayam Singh, Xiangru Tang, Leandro von Werra, Shayne Longpre. 2023.08

4. Magicoder: Source Code Is All You Need Preprint

   [Paper][Repo] Yuxiang Wei, Zhe Wang, Jiawei Liu, Yifeng Ding, Lingming Zhang 2023.12

▶️ Alignment with Feedback

 

1. CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning NeurIPS22

   [Paper] Hung Le, Yue Wang, Akhilesh Deepak Gotmare, Silvio Savarese, Steven C.H. Hoi. 2022.07

2. Execution-based Code Generation using Deep Reinforcement Learning TMLR23

   [Paper] Parshin Shojaee, Aneesh Jain, Sindhu Tipirneni, Chandan K. Reddy. 2023.01

3. RLTF: Reinforcement Learning from Unit Test Feedback Preprint

   [Paper] Jiate Liu, Yiqin Zhu, Kaiwen Xiao, Qiang Fu, Xiao Han, Wei Yang, Deheng Ye. 2023.07

4. PanGu-Coder2: Boosting Large Language Models for Code with Ranking Feedback Preprint

   [Paper] Bo Shen, Jiaxin Zhang, Taihong Chen, Daoguang Zan, Bing Geng, An Fu, Muhan Zeng, Ailun Yu, Jichuan Ji, Jingyang Zhao, Yuenan Guo, Qianxiang Wang. 2023.07

▶️ Prompting

 

1. CodeT: Code Generation with Generated Tests ICLR23

   [Paper] Bei Chen, Fengji Zhang, Anh Nguyen, Daoguang Zan, Zeqi Lin, Jian-Guang Lou, Weizhu Chen. 2022.07

2. Coder Reviewer Reranking for Code Generation ICML23

   [Paper] Tianyi Zhang, Tao Yu, Tatsunori B Hashimoto, Mike Lewis, Wen-tau Yih, Daniel Fried, Sida I Wang. 2022.11

3. LEVER: Learning to Verify Language-to-Code Generation with Execution ICML23

   [Paper] Ansong Ni, Srini Iyer, Dragomir Radev, Ves Stoyanov, Wen-tau Yih, Sida I. Wang, Xi Victoria Lin. 2023.02

4. Teaching Large Language Models to Self-Debug Preprint

   [Paper] Xinyun Chen, Maxwell Lin, Nathanael Schärli, Denny Zhou. 2023.06

5. Demystifying GPT Self-Repair for Code Generation Preprint

   [Paper] Theo X. Olausson, Jeevana Priya Inala, Chenglong Wang, Jianfeng Gao, Armando Solar-Lezama. 2023.06

6. SelfEvolve: A Code Evolution Framework via Large Language Models Preprint

   [Paper] Shuyang Jiang, Yuhao Wang, Yu Wang. 2023.06

▶️ Evaluation & Benchmark

 

1. Measuring Coding Challenge Competence With APPS NeurIPS21

   Named APPS

   [Paper][Repo] Dan Hendrycks, Steven Basart, Saurav Kadavath, Mantas Mazeika, Akul Arora, Ethan Guo, Collin Burns, Samir Puranik, Horace He, Dawn Song, Jacob Steinhardt. 2021.05

2. Program Synthesis with Large Language Models Preprint

   Named MBPP

   [Paper] Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, Charles Sutton. 2021.08

3. DS-1000: A Natural and Reliable Benchmark for Data Science Code Generation ICML23

   [Paper] Yuhang Lai, Chengxi Li, Yiming Wang, Tianyi Zhang, Ruiqi Zhong, Luke Zettlemoyer, Scott Wen-tau Yih, Daniel Fried, Sida Wang, Tao Yu. 2022.11

4. RepoBench: Benchmarking Repository-Level Code Auto-Completion Systems Preprint

   [Paper] Tianyang Liu, Canwen Xu, Julian McAuley. 2023.06

5. Can ChatGPT replace StackOverflow? A Study on Robustness and Reliability of Large Language Model Code Generation Preprint

   [Paper] Li Zhong, Zilong Wang. 2023.08

6. RepoCoder: Repository-Level Code Completion Through Iterative Retrieval and Generation EMNLP23

   [Paper] Fengji Zhang, Bei Chen, Yue Zhang, Jacky Keung, Jin Liu, Daoguang Zan, Yi Mao, Jian-Guang Lou, Weizhu Chen. 2023.10

7. CrossCodeEval: A Diverse and Multilingual Benchmark for Cross-File Code Completion Neurips23

   [Paper] Yangruibo Ding, Zijian Wang, Wasi Uddin Ahmad, Hantian Ding, Ming Tan, Nihal Jain, Murali Krishna Ramanathan. et al. 2023.11

8. SWE-bench: Can Language Models Resolve Real-World GitHub Issues? ICLR24

   [Paper] YCarlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, Karthik Narasimhan. 2023.10

9. DevBench: A Comprehensive Benchmark for Software Development Preprint

   [Paper][Repo] Yuxiang Wei, Zhe Wang, Jiawei Liu, Yifeng Ding, Lingming Zhang 2024.3

10. LongCoder: A Long-Range Pre-trained Language Model for Code Completion ICML23

    [Paper] Daya Guo, Canwen Xu, Nan Duan, Jian Yin, Julian McAuley. 2023.10

11. Coeditor: Leveraging Contextual Changes for Multi-round Code Auto-editing Preprint

    [Paper] Jiayi Wei, Greg Durrett, Isil Dillig. 2023.5

12. Automating Code Review Activities by Large-Scale Pre-training Preprint

    [Paper] JZhiyu Li, Shuai Lu, Daya Guo, Nan Duan, Shailesh Jannu, Grant Jenks, Deep Majumder, Jared Green, Alexey Svyatkovskiy, Shengyu Fu, Neel Sundaresan. 2022.10

▶️ Using LLMs while coding

 

1. Awesome-DevAI: A list of resources about using LLMs while building software Awesome

   [Repo] Ty Dunn, Nate Sesti. 2023.10