GV

GeezerVison

The World Wide Web, restored to 800 pixels of glory.

Home|About|Gaming cabinet|Archives|How it works|Starter sites

On this site

  • Sign in
  • ActionsCI/CD and automation
  • CodespacesCloud development environments
  • All Features
  • Documentation
  • Training
  • Certifications
  • Build with Copilot
  • Customize Copilot
  • Integrations
  • Trending repos
  • GitHub Sponsors
  • Developer articles
  • Enterprises

This view

  • Original page
  • New address

GeezerVison is showing github.com — GitHub - ise-uiuc/magicoder: [ICML'24] Magicoder: Empowering Code Generation with OSS-Instruct

BranchesTags
Open more actions menu

Latest commit

 

History

105 Commits
105 Commits

Folders and files

NameName
Last commit message
Last commit date
assets
assets
 
 
data
data
 
 
demo
demo
 
 
experiments
experiments
 
 
src/magicoder
src/magicoder
 
 
tests
tests
 
 
.gitignore
.gitignore
 
 
.pre-commit-config.yaml
.pre-commit-config.yaml
 
 
CITATION.cff
CITATION.cff
 
 
LICENSE
LICENSE
 
 
README-DEV.md
README-DEV.md
 
 
README.md
README.md
 
 
pdm.lock
pdm.lock
 
 
pyproject.toml
pyproject.toml
 
 

Repository files navigation

🎩 Magicoder: Source Code Is All You Need

Paper

🎩 Models | 📚 Dataset | 🚀 Quick Start | 👀 Demo | 📝 Citation | 🙏 Acknowledgements

We are thrilled that Magicoder and OSS-Instruct have inspired many amazing projects, including:

  • Google CodeGemma
  • OpenCodeInterpreter
  • RTLCoder
  • CodeUltraFeedback
  • speechless-coder-ds-6.7b

Contact: Yuxiang Wei, Zhe Wang, Yifeng Ding, Jiawei Liu, Lingming Zhang.

About

  • 🎩Magicoder is a model family empowered by 🪄OSS-Instruct, a novel approach to enlightening LLMs with open-source code snippets for generating low-bias and high-quality instruction data for code.
  • 🪄OSS-Instruct mitigates the inherent bias of the LLM-synthesized instruction data by empowering them with a wealth of open-source references to produce more diverse, realistic, and controllable data.

Important

  • Magicoder-S-DS-6.7B outperforms gpt-3.5-turbo-1106 and Gemini Ultra on HumanEval (76.8 vs. [72.6 and 74.4])!
  • Find more detailed comparisons with other SOTA models on the 🏆 EvalPlus Leaderboard 🏆!

Overview of OSS-Instruct Overview of Result

🎩 Models

Model Checkpoint Size HumanEval (+) MBPP (+) License
Magicoder-CL-7B 🤗 HF Link 7B 60.4 (55.5) 64.2 (52.6) Llama2
Magicoder-S-CL-7B 🤗 HF Link 7B 70.7 (66.5) 68.4 (56.6) Llama2
Magicoder-DS-6.7B 🤗 HF Link 6.7B 66.5 (60.4) 75.4 (61.9) DeepSeek
Magicoder-S-DS-6.7B 🤗 HF Link 6.7B 76.8 (70.7) 75.7 (64.4) DeepSeek

👀 Demo

Online Gradio Demo

Quickly try out our Magicoder Playground powered by gradio! Huge thanks to AK(@_akhaliq) and the Hugging Face team for their support!

Local Gradio Demo

We follow WizardCoder and provide the script to build a local demo server. You can launch your local gradio demo as following:

cd demo
CUDA_VISIBLE_DEVICES=0 python magicoder_demo.py \
   --base_model "ise-uiuc/Magicoder-S-DS-6.7B" \
   --device "cuda:0" \
   --port 8080

📚 Dataset

  • Magicoder-OSS-Instruct-75K: generated through OSS-Instruct using gpt-3.5-turbo-1106 and used to train both Magicoder and Magicoder-S series.
  • Magicoder-Evol-Instruct-110K: decontaminated and redistributed from theblackcat102/evol-codealpaca-v1, used to further finetune Magicoder series and obtain Magicoder-S models.

🚀 Quick Start

from transformers import pipeline
import torch

MAGICODER_PROMPT = """You are an exceptionally intelligent coding assistant that consistently delivers accurate and reliable responses to user instructions.

@@ Instruction
{instruction}

@@ Response
"""

instruction = "Implement a high-level API for a TODO list application. The API takes as input an operation request and updates the TODO list in place. If the request is invalid, raise an exception."

prompt = MAGICODER_PROMPT.format(instruction=instruction)
generator = pipeline(
    model="ise-uiuc/Magicoder-S-DS-6.7B",
    task="text-generation",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
result = generator(prompt, max_length=2048, num_return_sequences=1, temperature=0.0)
print(result[0]["generated_text"])

This code snippet will generate the following output:

Here is a simple Python implementation of a TODO list API:

```python
class TodoList:
    def __init__(self):
        self.todo_list = []

    def add_task(self, task):
        if not isinstance(task, str):
            raise ValueError("Task must be a string")
        self.todo_list.append(task)

    def remove_task(self, task):
        if task not in self.todo_list:
            raise ValueError("Task not found in the list")
        self.todo_list.remove(task)

    def get_tasks(self):
        return self.todo_list

    def update_task(self, old_task, new_task):
        if old_task not in self.todo_list:
            raise ValueError("Old task not found in the list")
        if not isinstance(new_task, str):
            raise ValueError("New task must be a string")
        index = self.todo_list.index(old_task)
        self.todo_list[index] = new_task

    def clear_list(self):
        self.todo_list = []
```

This API allows you to add tasks, remove tasks, get all tasks, update tasks, and clear the list. It also raises exceptions for invalid operations.

You can use this API like this:

```python
todo = TodoList()
todo.add_task("Buy groceries")
todo.add_task("Finish project")
print(todo.get_tasks())  # Output: ['Buy groceries', 'Finish project']
todo.update_task("Buy groceries", "Buy fruits")
print(todo.get_tasks())  # Output: ['Buy fruits', 'Finish project']
todo.remove_task("Finish project")
print(todo.get_tasks())  # Output: ['Buy fruits']
todo.clear_list()
print(todo.get_tasks())  # Output: []
```

📝 Citation

@InProceedings{wei2024magicoder,
  title = 	 {Magicoder: Empowering Code Generation with {OSS}-Instruct},
  author =       {Wei, Yuxiang and Wang, Zhe and Liu, Jiawei and Ding, Yifeng and Zhang, Lingming},
  booktitle = 	 {Proceedings of the 41st International Conference on Machine Learning},
  pages = 	 {52632--52657},
  year = 	 {2024},
  volume = 	 {235},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {21--27 Jul},
  publisher =    {PMLR},
  pdf = 	 {https://raw.githubusercontent.com/mlresearch/v235/main/assets/wei24h/wei24h.pdf},
  url = 	 {https://proceedings.mlr.press/v235/wei24h.html}
}

🙏 Acknowledgements

We thank AK(@_akhaliq) and the Hugging Face team for their support in the Magicoder Playground! We also thank the following amazing projects that truly inspired us:

  • WizardCoder: Evol-Instruct
  • DeepSeek-Coder: Base model for Magicoder-DS
  • CodeLlama: Base model for Magicoder-CL
  • StarCoder: Data decontamination

⚠️ Important Note

  • Bias, Risks, and Limitations: Magicoders may sometimes make errors, produce misleading contents, or struggle to manage tasks that are not related to coding.

  • Usage: Magicoder models are trained on the synthetic data generated by OpenAI models. Please pay attention to OpenAI's terms of use when using the models and the datasets. Magicoders will not compete with any OpenAI's commercial product.

⭐️ Star History

Star History Chart

About

[ICML'24] Magicoder: Empowering Code Generation with OSS-Instruct

proceedings.mlr.press/v235/wei24h.html

Topics

ai4codelarge-language-modelsllmllm4code

Resources

Readme
MIT license
Activity
Custom properties

Stars

2.1k stars

Watchers

24 watching

Forks

169 forks
Report repository

Used by

Contributors

Languages

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.