The new release also introduces image-in/text-out functionality and improved foreign language capabilities, supporting a dozen languages alongside English. Licensed under the Apache 2 open-source license, IBM provides transparency by disclosing training datasets and offering comprehensive documentation. Granite 3.1 models are available on platforms like IBM's Watsonx, Google Vertex AI, and Hugging Face, aiming to accelerate AI adoption in enterprises. By combining these models with proprietary data, businesses can achieve task-specific performance at a lower cost, positioning Granite 3.1 as a powerful and trustworthy AI tool for enterprise innovation.
Key takeaways:
- IBM's Granite 3.1 models feature a 128K token context window, significantly enhancing their ability to process and understand large amounts of text for comprehensive analysis and generation tasks.
- The Granite 8B Instruct model reportedly outperforms competitors like Google Gemma 2 and Meta Llama 3.1 on HuggingFace's OpenLLM Leaderboard benchmarks.
- Granite 3.1 introduces image-in/text-out functionality and improved foreign language capabilities, supporting a dozen languages, making it suitable for international business applications.
- IBM's Granite 3.1 models are available on various platforms, including IBM's Watsonx, Google Vertex AI, and Hugging Face, and are designed to accelerate AI adoption in enterprise settings.