Magicoder is a Python-based code generation model family using OSS-Instruct, with multiple 7B models and datasets for training; it is MIT-licensed and hosted on HuggingFace.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Magicoder is a model family empowered by OSS-Instruct, aiming to generate code with low bias by using open-source code references to produce instruction data for code generation. The project provides multiple 7B-class models and datasets used for training Magicoder series.
How it works
The README describes OSS-Instruct as a method to mitigate bias in LLM-instruct data by leveraging open-source references to create diverse, realistic instruction data for code tasks. It also shows several model checkpoints with associated evaluation metrics on code benchmarks (e.g., HumanEval, MBPP).
Getting started
The Quick Start section includes a Python code snippet to run a generator using a specific model:
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 demonstrates how to instantiate the model for text-generation tasks using the HuggingFace transformers pipeline.
Recent releases
The Releases section lists latest 0 with no entries under RELEASES (latest 0): none.
Traction
Stars: 2095 Forks: 171 Open issues: 4
Behind the repo
No dedicated startup/company link is provided in the facts.
Caveats
License: MIT. Created 2023-11-10, last_push 2024-11-01. Language: Python. License is MIT as shown in the README. The dataset references and model links are provided, including license notes for specific model checkpoints.