ibm-granite/granite-3.2-8b-instruct-preview
Granite-3.2-8B-Instruct-Preview is an 8 billion parameter long-context instruction-tuned model developed by IBM, building upon Granite-3.1-8B-Instruct. It is specifically fine-tuned for enhanced reasoning capabilities, utilizing a mix of permissively licensed open-source and internally generated synthetic data. This model supports a 32768 token context length and offers controllable 'thinking' functionality, making it suitable for general instructions and AI assistant development in various domains, including business applications.
Loading preview...
Overview
Granite-3.2-8B-Instruct-Preview is an 8 billion parameter, long-context instruction-tuned model from IBM, released as an early preview. It is an evolution of Granite-3.1-8B-Instruct, with a primary focus on enhancing reasoning capabilities. The model was trained using a combination of permissively licensed open-source datasets and proprietary synthetic data specifically designed for complex reasoning tasks. A key feature is its controllable 'thinking' capability, which can be activated as needed.
Key Capabilities
- Enhanced Reasoning: Fine-tuned with synthetic data to improve problem-solving and logical deduction.
- Long Context Window: Supports a 32768 token context length, enabling processing of extensive documents and conversations.
- Multilingual Support: Designed for English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, and Chinese, with potential for fine-tuning in other languages.
- Versatile Applications: Capable of summarization, text classification, extraction, question-answering, RAG, code-related tasks, function-calling, and multilingual dialogue.
Performance Highlights
Compared to its predecessor, Granite-3.1-8B-Instruct, and other 8B-class models like Llama-3.1-8B-Instruct and Qwen-2.5-7B-Instruct, Granite-3.2-8B-Instruct-Preview shows significant improvements in specific benchmarks. Notably, it achieves 55.23 on ArenaHard and 61.16 on Alpaca-Eval-2, indicating strong performance in instruction following and helpfulness.
Intended Use Cases
This model is designed for general instruction response and building AI assistants across various domains, including business applications. Its enhanced reasoning and long-context handling make it particularly suitable for tasks requiring deep understanding and complex problem-solving.