This session involves practical and theoretical demonstrations of how to train and deploy your own large language model (LLM) on Android.

The session tackles ML problem framing such as text generation using a fine-tuned pre-trained GPT-2 model.

We explore the use of datasets using Dataset Search to prepare data for our LLM. We create a TensorFlow dataset and run preprocessing on it. The tf.data API helps to map the gpt2_preprocessor against the training dataset.

We train an LLM using pre-trained models and deploy them to Android using TensorFlow Serving.

This session assumes that the learner is aware of mobile technologies like Android and Flutter and also aware of Natural Language Processing.