Granite-3.2-2B-Instruct
| Property | Value |
|---|---|
| Parameter Count | 2 Billion |
| Release Date | February 26th, 2025 |
| License | Apache 2.0 |
| Developer | IBM Granite Team |
| Model URL | huggingface.co/ibm-granite/granite-3.2-2b-instruct |
What is granite-3.2-2b-instruct?
Granite-3.2-2B-Instruct is IBM's latest 2-billion parameter language model, specifically designed for enhanced reasoning and instruction-following capabilities. Built upon its predecessor Granite-3.1-2B-Instruct, this model introduces controllable thinking abilities and supports 12 languages, making it versatile for various business applications and AI assistance tasks.
Implementation Details
The model is trained using IBM's Blue Vela supercomputing cluster with NVIDIA H100 GPUs, utilizing a combination of permissively licensed open-source datasets and internally generated synthetic data. It implements a unique controllable thinking mechanism that can be toggled based on task requirements, offering flexibility in reasoning depth.
- Supports multiple languages including English, German, Spanish, French, Japanese, and others
- Implements controllable thinking through a 'thinking' parameter in the generation process
- Trained on specialized synthetic data for enhanced reasoning capabilities
- Optimized for long-context tasks and document processing
Core Capabilities
- Advanced thinking and reasoning tasks
- Text summarization and classification
- Information extraction and question-answering
- Retrieval Augmented Generation (RAG)
- Code-related tasks and function calling
- Long-context document processing
- Multilingual dialogue support
Frequently Asked Questions
Q: What makes this model unique?
The model's distinguishing feature is its controllable thinking capability, allowing users to activate advanced reasoning when needed while maintaining efficiency for simpler tasks. It also shows impressive performance on benchmark tests, particularly in ArenaHard and Alpaca-Eval-2 compared to its predecessor.
Q: What are the recommended use cases?
The model excels in business applications requiring reasoning, document processing, and multilingual support. It's particularly suited for tasks like document summarization, complex question-answering, and code-related applications, with benchmark scores showing strong performance in human evaluation and truthfulness tests.
