What is Granite 3.1
Granite 3.1 is a new generation language model launched by IBM, featuring strong performance and longer context handling capabilities. The Granite 3.1 model family has 4 different sizes and 2 architectures: dense models with 2B and 8B parameter models, trained on a total of 12 trillion tokens. Mixture of Experts MoE models: sparse 1B and 3B MoE models, with 400M and 800M activated parameters respectively, trained on a total of 10 trillion tokens. Granite 3.1 supports a context length of 128K tokens and can handle complex text and tasks. Granite 3.1 introduces a new embedding model, covering 12 languages and enhancing multilingual support. The Granite Guardian 3.1 model has function-calling hallucination detection capabilities, improving control and observability over tool calls.
Main features of Granite 3.1
- Extended context window: The context window is extended to 128K tokens, supporting the model in processing larger inputs, conducting longer continuous interactions, and incorporating more information into each output.
- New embedding models: A series of new retrieval-optimized Granite Embedding models are introduced, offering different sizes from 30M to 278M parameters, with multilingual support for 12 different languages.
- Function-calling hallucination detection: The Guardian 3.1 8B and 2B models add function-calling hallucination detection capabilities, enhancing control and observability over agents making tool calls.
Technical principles of Granite 3.1
- Large Language Model (LLM): Built on large language models, trained through deep learning and natural language processing technologies to understand and generate natural language text.
- Transformer Architecture: Uses Transformer architecture to capture long-distance dependencies in text.
- Context Window: The context window is extended to 128K tokens, allowing the model to consider more information at the same time and handle longer text and more complex tasks.
- Multilingual Support: Trains the model to recognize and understand multiple languages, supporting text processing in 12 different languages.
Project address of Granite 3.1
- Project official website：ibm-granite-3-1
- GitHub repository：https://github.com/ibm-granite/granite-3.1
- HuggingFace model library：https://huggingface.co/collections/ibm-granite/granite-31
Application scenarios of Granite 3.1
- Customer service automation: Build chatbots and virtual assistants, automate customer inquiries and support services, and improve response speed and efficiency.
- Content creation and editing: Help content creators generate, edit, and optimize text content, including articles, reports, and marketing materials.
- Enterprise search and knowledge management: Based on semantic search and vector search technologies, improve enterprise internal knowledge retrieval systems and help employees quickly find the information they need.
- Language translation and localization: Translation services and localization projects for multinational companies.
- Risk management and compliance checks: In the financial and legal fields, help analyze and understand large numbers of documents, ensuring compliance and risk management.