OpenAI announced on Thursday that it will roll out new features to improve the fine-tuning application programming interface (API) by developers with greater control over their fine-tuning tasks.
The latest updates will include saving a full model checkpoint at every training stage to minimize the necessity for retraining, a new side-by-side Playground UI for comparing various models' quality and performance as well as integration support with other platforms. Furthermore, metrics will be computed over the validation dataset at the conclusion of each training phase, and developers will be able to adjust hyperparameters directly from the Dashboard, the company shared in the statement.
In addition, OpenAI also introduced an "assisted fine-tuning offering" as part of the Custom Model program, which was previously launched last November. This feature aims to aid organizations requiring assistance in establishing effective training data pipelines, evaluation systems, and customized parameters and methods to optimize model performance for their specific use case or task.