Saikrishna2511/qwen-multitask
Saikrishna2511/qwen-multitask is a 0.5 billion parameter Qwen2.5-Coder-0.5B-Instruct checkpoint, fine-tuned for multi-task code generation and documentation. It excels at natural language to Python code generation, Java to Python translation, and Python code documentation. With a 32768 token context length, this model is optimized for various programming-related tasks.
Loading preview...
Model Overview
Saikrishna2511/qwen-multitask is a specialized 0.5 billion parameter model based on the Qwen2.5-Coder-0.5B-Instruct architecture. It has been multi-task fine-tuned using LoRA to handle several programming-related tasks efficiently, making it a versatile tool for developers.
Key Capabilities
This single checkpoint is designed to perform three distinct tasks, each activated by a specific prompt prefix:
- Natural Language to Python (
nl2py): Converts natural language descriptions into Python code. - Java to Python Translation (
java2py): Translates Java code snippets into equivalent Python code. - Code to Documentation (
code2doc): Generates documentation for given Python code.
Training Details
The model underwent a two-stage LoRA fine-tuning process:
- Stage 1: Java to Python translation using the AVATAR-TC dataset.
- Stage 2: Multi-task training on NL2Py, Code2Doc, code comments, and a replay of Java2Py tasks.
Limitations
Due to its compact 0.5B parameter size, the model's quality can vary depending on the complexity of the task and input. While primarily trained on Python, Java translation quality is dependent on the training coverage. It is not recommended for production use without further rigorous evaluation.