What Open Source Means in LLMs — and the IBM Granite Advantages

When choosing the right large language model, businesses have a lot to consider: from selecting the right model type to balancing model size with business needs. One key factor in the decision-making process is how these models are licensed. While proprietary models are managed under closed licenses, open-source LLMs offer unique benefits—but even open-source models can vary greatly depending on their specific licenses.

In this article, I’ll break down what open source means in the LLM world and explain IBM’s approach with its Granite models, which blend open-source access with enterprise-grade dependability.

Proprietary vs. Open-Source LLMs

Large language models fall into two main categories: proprietary and open source. Proprietary models, developed and maintained by specific companies, are usually restricted in terms of access, usage, and modification. Open-source models, by contrast, are freely available to the public, allowing anyone to customize them to meet specific needs or integrate them into specialized applications—offering greater transparency and flexibility.

In the context of LLMs, open source can be an asset for businesses aiming to build customized solutions while maintaining some level of control over the technology that powers them.

?

Benefits of Open Source

Open-source LLMs offer businesses several unique advantages:

• Transparency: Open-source models allow businesses to examine how the model operates "under the hood," offering visibility into its architecture and inner workings. This level of transparency provides valuable insight into the model's processes and enables thorough audits of its behavior when needed.
• Flexibility and Customization: Open-source models can be adapted to fit specific business needs, customized with proprietary data, or integrated with other tools to create tailored applications that stand apart.
• Community and Collaboration: Open-source models benefit from contributions from global developers and researchers, accelerating innovation and expanding the support available for these models.

?

Open-Source Licenses: MIT, Apache 2.0, and GPL

Not all open-source licenses are created equal. Here’s a quick look at three ones and what they mean:

1. MIT License: A highly permissive license that allows anyone to use, modify, and distribute software freely—even for commercial purposes—as long as the original license is included.
2. Apache License 2.0: Similar to the MIT License but with added protections against patent claims, which can be crucial for businesses wanting to safeguard their IP.
3. GNU General Public License (GPL): A more restrictive license that requires any derivative works to also be open-sourced under the same license. This means companies need to make their modifications public if they distribute the software, which may not suit all business needs.

?

IBM Granite Model Approach: Enterprise-Ready Open Source

IBM’s Granite models exemplify how open-source LLMs can be tailored to meet the demands of the enterprise. Developed to support business-grade applications, these models are trained on curated, industry-relevant datasets spanning five essential domains: internet, academic, code, legal, and finance. This focused training enables Granite models to generate text and insights that align with professional, enterprise-level needs. Additionally, IBM prioritizes trust and transparency, ensuring Granite models are interpretable, auditable, and robust, all aligned with IBM’s ethical AI principles.

Granite Licensing: Open Source Meets Enterprise Security

IBM’s Granite models and the watsonx platform take an ecosystem approach, enabling businesses to not only deploy AI but also to become AI value creators. Granite models are released under the Apache 2.0 license, which allows for broad commercial use and provides developers the flexibility to fine-tune the models. By integrating proprietary data with Granite’s foundation, businesses can develop AI models that reflect their unique industry context and competitive strengths.

IBM’s decision to release Granite under the Apache 2.0 license aligns with its vision of making AI accessible to a wide range of developers. The permissive nature of this license, combined with its protections against patent claims, makes Granite a robust choice for companies looking to leverage AI without facing entry barriers. These models are also accessible on Hugging Face, a popular platform that encourages community engagement and innovation.

IP Indemnity: A Unique Security for IBM Clients

IBM goes a step further with its Intellectual Property Indemnity for Granite models. This protection means that clients using Granite models can rely on IBM for defense against IP claims related to the models. This indemnity is a crucial safeguard for companies looking to use proprietary data with Granite models. In an era where data is the source of competitive advantage, this security allows businesses to confidently develop AI applications without IP concerns.

Open-Source Models: Free, but Not Costless

Finally, it’s essential to understand that while open-source LLMs like Granite, LLaMA, and Falcon are freely accessible, that doesn’t mean they’re entirely cost-free. Here’s why:

1. Infrastructure and Scaling Costs: Running large models requires significant computing resources, which often means investing in powerful hardware or cloud storage. Many companies rely on third-party providers to host and manage these models, incurring costs even when the models themselves are free.
2. Customization and Maintenance: While open-source models offer a strong foundation, they often need customization to suit specific use cases. Tailoring and maintaining these models requires skilled teams, whether in-house or outsourced, which translates into expenses for development and support.
3. Support and Security: Managed services or third-party support options can provide critical reliability and security updates. For many businesses, paying for this support is essential to maintain operational stability and safeguard sensitive data.

?

IBM’s Granite models strike a balance between the openness of open-source licensing and the reliability enterprises need in the AI landscape. With Granite and watsonx, IBM enables businesses to confidently deploy custom AI models, backed by comprehensive IP protections and designed to harness proprietary data as a competitive advantage.