Developing and fine-tuning language model systems is a challenging process that usually consumes a significant amount of time and resources even for tech giants like Google and Meta. It involves an iterative process of supervised fine-tuning, aligning with human preferences, distillation, and continuous adjustment until a certain quality threshold is met. This process can take 12 to 18 months, which is often not feasible for many businesses.
The startup Automorphic aims to address these issues, providing a solution that allows programmers to easily develop and refine personalized models. In just a few minutes, a model can be fine-tuned and put into production using the company’s language model improvement platform. Users input their raw text data, initiate the first fine-tuning run, and then make necessary adjustments.
The platform aligns with OpenAI’s API, implying that users only have to change one line to use Automorphic’s endpoint. Users can then train adapters with more data and combine them, progressively refining their models. Additionally, Automorphic’s hub facilitates testing and publishing of customized models.
One of Automorphic’s main products is Conduit, a platform that uses fine-tuning to incorporate knowledge into language models, overcoming the limitations of context windows. It enables the creation of behavioral and knowledge adapters, which can be interchangeably used as needed. Conduit allows a faster transition of models into production through an iterative process and human-in-the-loop input. It is compatible with the OpenAI API, making it easy to use without altering current codes. Conduit also enhances datasets to improve models gradually.
With Conduit, users can swiftly load and stack fine-tuned adapters. This allows them to focus on giving feedback and making tweaks, independently of performance and deployment issues.
In essence, Automorphic allows developers to rapidly transform raw data into a bespoke language model that can be deployed and improved over time. Using Automorphic’s product, Conduit, developers can create domain-specific language models in a cost-effective and time-efficient manner. Therefore, Automorphic could potentially revolutionize the process of building and improving language models, making it accessible for more businesses.
