🎉 You are looking at Kite 1.5, which is using an already randomized model (with the same config) during training, and has been trained with significantly higher steps!

Kite is a small, trained, 1 million parameter language model, without any special optimizations.

It was trained on this dataset using 20000 steps, 1 epoch, 1 batch size, and the GPT-2 tokenizer.

Due to its size, the model is not suitable for production workloads.