Enterprise Large Language Models (LLMs)

Enterprise Large Language Models (LLMs) are LLMs that can be managed by enterprises (instead of LLM providers). They can be deployed on-prem and fine-tuned by enterprises.

These are language models that have been trained on large amounts of text and have diverse text processing capabilities. For example, they can be used to power AI-driven chatbots, extract insights from large volumes of text data and develop knowledge management systems.  Enterprises can use these foundation models as their base foundation model which they can finetune with their own data. If they prefer, finetuning can be done on-premise without exposing confidential data to 3rd party services. 

To be included in this list, the model needs to be

• Provided with its pre-trained weights so users can modify the model and examine how it works.
• Provided with a license that permits commercial use. Commercial use may have limitations. For example, Meta's models require a license agreement for companies with 700 million active users.
• Models with licenses restricting only harmful use of AI like the RAIL license are also included in this list.
• Deployable in an on-prem or cloud setting, enabling businesses to not share their data with 3rd parties and reduce their data attack surface.

It is also important for the model to be either

• trained using a set of data that does not include any copyright protected data or
• providing indemnity against copyright-related lawsuits regarding the LLM's output.

However, few providers fulfill this last criteria. Therefore, we did not make it a requirement to join this list to give enterprises more options.