🎩 Models | 📚 Dataset | 🚀 Quick Start | 👀 Demo | 📝 Citation | 🙏 Acknowledgements
We are thrilled that Magicoder and OSS-Instruct have inspired many amazing projects, including:
Contact: Yuxiang Wei, Zhe Wang, Yifeng Ding, Jiawei Liu, Lingming Zhang.
- 🎩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-1106and Gemini Ultra on HumanEval (76.8 vs. [72.6 and 74.4])! - Find more detailed comparisons with other SOTA models on the 🏆 EvalPlus Leaderboard 🏆!
| 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 |
Quickly try out our Magicoder Playground powered by gradio! Huge thanks to AK(@_akhaliq) and the Hugging Face team for their support!
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- Magicoder-OSS-Instruct-75K: generated through OSS-Instruct using
gpt-3.5-turbo-1106and 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.
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: []
```
@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}
}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
-
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.