Granite-34B-Code-Instruct-8K
| Property | Value |
|---|---|
| Developer | IBM Research |
| Parameters | 34 Billion |
| Release Date | May 6th, 2024 |
| License | Apache 2.0 |
| Paper | Granite Code Models: A Family of Open Foundation Models for Code Intelligence |
What is granite-34b-code-instruct-8k?
Granite-34B-Code-Instruct-8K is an advanced language model specifically designed for code intelligence and programming tasks. Built on the foundation of Granite-34B-Code-Base, this model has been fine-tuned using a carefully curated collection of permissively licensed instruction data to enhance its code understanding and generation capabilities.
Implementation Details
The model was trained on IBM's supercomputing clusters (Vela and Blue Vela) using NVIDIA A100 and H100 GPUs. It incorporates diverse training data including code commits from CommitPackFT, mathematical datasets (MathInstruct and MetaMathQA), and specialized code instruction datasets like Glaive-Code-Assistant-v3.
- 8K context window for handling larger code segments
- Support for 92 programming languages
- Integrated with popular frameworks via HuggingFace transformers
- Optimized for both CPU and GPU deployment
Core Capabilities
- Advanced code generation and completion
- Logical reasoning and problem-solving
- Mathematical computation assistance
- API calling and function implementation
- SQL query generation
Frequently Asked Questions
Q: What makes this model unique?
The model's distinctive feature is its specialized training on high-quality, permissively licensed code instruction data, combined with its large 34B parameter size and 8K context window, making it particularly effective for complex coding tasks.
Q: What are the recommended use cases?
The model excels in building coding assistants, handling programming tasks across multiple languages, mathematical problem-solving, and API integration scenarios. It's particularly suited for applications requiring deep code understanding and generation capabilities.
