dharun2049/Kaveri-4B
Kaveri-4B is a 4.5 billion parameter language model developed by dharun2049, based on Qwen3.5-4B. It is specifically fine-tuned for software engineering tasks, excelling in algorithmic problem solving, competitive programming, and Python code generation. This compact model is optimized for code generation and algorithmic reasoning, making it suitable for programming assistance and coding-model research.
Loading preview...
Kaveri-4B: A Compact Coding-Focused LLM
Kaveri-4B is a 4.5 billion parameter language model developed by dharun2049, built upon the Qwen3.5-4B base architecture. This model is specifically fine-tuned to excel in various software engineering domains, with a strong emphasis on code generation and algorithmic problem-solving.
Key Capabilities
- Code Generation: Highly optimized for generating Python code.
- Algorithmic Reasoning: Designed to handle algorithmic problems and competitive programming challenges.
- Software Engineering: Fine-tuned for general software development tasks.
Training Details
The model was fine-tuned using 16-bit LoRA with a rank of 16 and an alpha of 16. It underwent 500 training steps with an effective batch size of 8. The training data consisted of execution-filtered NVIDIA OpenCodeInstruct, with a quality threshold of 0.80, ensuring a focus on high-quality code and reasoning examples.
Intended Uses
Kaveri-4B is ideal for developers and researchers focused on:
- Generating code snippets and functions.
- Implementing algorithms efficiently.
- Assisting in competitive programming scenarios.
- General programming assistance.
- Research and development in coding models.