# Trainable on a single GPU

TinyMistral-248M is a small language model based on pre-training from the Mistral 7B model, with parameters scaled down to approximately 248 million, primarily used for fine-tuning downstream tasks.