Graft provides a number of standard large language models which are trained on a large corpora of data and are suitable for a wide range of use cases; there may be situations where tailoring the model to your specific use case context can improve results, for example if you are building a semantic search model for pharmacological terms. In this instance using some of your own data to update the model, known as pre-training, can be an effective away to build a custom trunk model without the extensive effort of building a model from scratch.
To pre-train a model you will need your own data connected as a data source in Graft. For details on how to create your own Custom pretrained model please see the Custom Model documentation.
Custom Model Scope
Custom Trunk Models are available within the Project they are created in, at this time it is not possible to share a new model across multiple projects.