RedHatAI/granite-3.2-2b-instruct
RedHatAI/granite-3.2-2b-instruct is a 2-billion-parameter instruction-tuned language model developed by the Granite Team at IBM, building upon Granite-3.1-2B-Instruct. It features a 32768-token context length and is specifically fine-tuned using synthetic data to enhance its reasoning capabilities, allowing for controllable 'thinking' processes. This model is designed for general instruction-following, multilingual dialog, and long-context tasks, making it suitable for integration into AI assistants for business applications.
Loading preview...
Overview
Granite-3.2-2B-Instruct is a 2-billion-parameter, long-context AI model developed by the Granite Team at IBM, released on February 26th, 2025, under the Apache 2.0 License. This model is an evolution of Granite-3.1-2B-Instruct, specifically fine-tuned to enhance its reasoning capabilities. It leverages a combination of permissively licensed open-source datasets and internally generated synthetic data tailored for complex reasoning tasks. A key feature is the controllability of its thinking capability, which can be activated only when required, as demonstrated in the provided acid solution problem example.
Key Capabilities
- Enhanced Reasoning: Fine-tuned with synthetic data for improved thinking capabilities.
- Long-Context Processing: Handles tasks requiring a 32768-token context length, such as long document summarization and QA.
- Multilingual Support: Supports English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese, with potential for further fine-tuning in other languages.
- General Instruction Following: Designed for a wide range of instruction-following tasks.
- Diverse Applications: Includes summarization, text classification, text extraction, question-answering, Retrieval Augmented Generation (RAG), code-related tasks, and function-calling.
Good For
- AI Assistants: Ideal for integration into AI assistants, particularly in business applications.
- Complex Problem Solving: Suitable for tasks requiring explicit reasoning and thought processes.
- Multilingual Deployments: Effective in environments requiring support for multiple languages.
- Long Document Analysis: Excels in scenarios involving extensive text, such as legal documents or meeting transcripts.
- Developers seeking controllable reasoning: The
thinking=Trueparameter allows for explicit control over the model's reasoning output, which can be valuable for debugging or understanding its decision-making process.