EXTRACTUM.IO BLOG

Understanding Permissive Licenses for Large Language Models (LLMs)

The booming artificial intelligence (AI) industry has revolutionized the way we interact with data. An integral part of this innovation is the advent of Large Language Models (LLMs), capable of mimicking human-like text generation. However, the effective and responsible usage of these models raises critical questions around licensing.

Licensing LLMs is a complex issue due to the unique characteristics of these models. Unlike traditional software, LLMs are not strictly “code” but are trained on large data sets. Hence, the application of traditional open-source licenses is not always straightforward.

To strike a balance between the open usage and protection of the original work, permissive licenses are widely used in the AI community. These licenses include Apache-2.0, BSD, BSD-2-Clause, BSD-3-Clause, MIT, and various Creative Commons licenses, each with unique stipulations.

Navigating through the complexities of various licenses can be daunting. To help clear the fog around these licenses and their stipulations, check the following comprehensive comparison table. This table outlines key aspects of each license, presenting a snapshot of their requirements for both users and authors, as well as highlighting potential pitfalls for commercial use. This comparative framework provides an easier way to weigh up the different licenses and find the one that suits your specific needs. Now, let’s take a look at this table.
Apache-2.0 (Apache License 2.0)

Requirements for the user:

  • Any significant changes made to the Large Language Model (LLM) must be documented.
  • Include a copy of the original Apache License in any use of the LLM.
  • Existing copyright, patent, trademark, and attribution notices must be maintained.

Requirements for the author:

  • Provide a copy of the Apache License.
  • Provide a notice file with the above-mentioned information.

Potential pitfalls for commercial use:

  • One unique aspect of the Apache License is that it includes an express grant of patent rights from contributors to users. Businesses need to be cautious about this clause if they are also developing their own proprietary software.

BSD (Berkeley Software Distribution license)

Requirements for the user:

  • Use of the LLM must keep the original copyright notice, list of conditions, and disclaimer.

Requirements for the author:

  • The license text must be included in the distribution of the LLM.

Potential pitfalls for commercial use:

  • The original BSD License has an “advertising clause” that requires all advertising materials mentioning the LLM to display an acknowledgement. This can be burdensome for businesses.

BSD-2-Clause (BSD 2-Clause “Simplified” License)

Requirements for the user:

  • Use of the LLM must keep the original copyright notice, list of conditions, and disclaimer.

Requirements for the author:

  • The license text must be included in the distribution of the LLM.

Potential pitfalls for commercial use:

  • This is a very permissive license, with few conditions, making it straightforward for commercial use.

BSD-3-Clause (BSD 3-Clause “New” or “Revised” License)

Requirements for the user:

  • Use of the LLM must keep the original copyright notice, list of conditions, and disclaimer.
  • Cannot use the name of the original author or contributors to endorse or promote products derived from the LLM without permission.

Requirements for the author:

  • The license text must be included in the distribution of the LLM.

Potential pitfalls for commercial use:

  • The restriction on the use of the names of the original author or contributors for promotional purposes can pose a limitation in some commercial contexts.

CC-BY-2.0 (Creative Commons Attribution 2.0)

Requirements for the user:

  • Must give appropriate credit, provide a link to the license, and indicate if changes were made to the LLM.

Requirements for the author:

  • The license text must be included in the distribution of the LLM.

Potential pitfalls for commercial use:

  • The requirement to provide credit and indicate changes may be burdensome in some commercial contexts, especially for large-scale uses of the LLM.

Despite the permissiveness of these licenses, some impose requirements that may affect the utilization of LLMs in a commercial context. For instance, Apache-2.0 and BSD-3-Clause licenses stipulate that any modifications to the model must be documented, a condition that might not suit certain businesses.

Moreover, Creative Commons licenses (CC-BY-2.0, 3.0, 4.0) mandate giving appropriate credit, providing a link to the license, and indicating if changes were made. These requirements might be burdensome in large-scale commercial applications of LLMs.

However, the MIT and BSD-2-Clause licenses are often preferred for commercial use due to their simplicity and minimal requirements. They only mandate the retention of copyright and permission notices in the distribution of the model.

The right licensing choice depends on the nature of the project, the scale of use, and the flexibility required by the user. It’s essential for developers and businesses to carefully consider the implications of each license to ensure they can meet the required obligations. A well-considered license can open up avenues for innovation while ensuring respect for the original creators’ rights.

If you’re diving into the world of Large Language Models and feeling a bit overwhelmed by all the different options and licenses, we’ve got a solution for you — the LLM Explorer. This handy service facilitates the hassle-free exploration of thousands of LLMs, breaking down all the important details for you.
From understanding licensing details to picking out the best model for your local inference needs based on license type and other factors, the LLM Explorer does the heavy lifting. It even lets you compare similar LLMs, so you can make the best choice for your project.

At the end of the day, choosing the right license depends on what your project needs. So why not make things easier with the LLM Explorer? Give it a try and take the guesswork out of using Large Language Models.
Made on
Tilda