Granite 3.3

The IBM Granite 2B and 8B models are 128K context length language models that have been fine-tuned for improved reasoning and instruction-following capabilities. These models deliver significant gains on benchmarks for measuring generic performance including AlpacaEval-2.0 and Arena-Hard, and improvements in mathematics, coding, and instruction following. They also supports Fill-in-the-Middle (FIM) for code completion tasks and structured reasoning.

Supported Languages

English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese. However, users may finetune this Granite model for languages beyond these 12 languages. Intended Use

These models are designed to handle general instruction-following tasks and can be integrated into AI assistants across various domains, including business applications.

Capabilities

• Thinking
• Summarization
• Text classification
• Text extraction
• Question-answering
• Retrieval Augmented Generation (RAG)
• Code related tasks
• Function-calling tasks
• Multilingual dialog use cases
• Fill-in-the-middle
• Long-context tasks including long document/meeting summarization, long document QA, etc.

Thinking

To enable thinking, add a message with “role”: “control” and set “content” to “thinking”. For example:

{
    "messages": [
        {"role": "control", "content": "thinking"},
        {"role": "user", "content": "How do I get to the airport if my car won't start?"}
    ]
}

Learn more

• Developers: IBM Research
• Website: Granite Docs
• Release Date: April 16th, 2025
• License: Apache 2.0

Credits

In this model 8B parameter model, I’ve modified the modelfile to enable thinking always, so it doesn’t need to be explicitly mentioned by the user to use it. I have only made that change, the model is created and fine-tuned by IBM, so the full credits go to them.😇