---
license: cc0-1.0
datasets:
- openbmb/Ultra-FineWeb-L3
language:
- en
pipeline_tag: text-generation
library_name: transformers
---

# Kite

🎉 You are looking at Kite 6.2, which uses a different and cleaner dataset!

Kite is a small, trained, 18 million parameter language model.

## Training

It was trained on a tokenized and truncated version of the "multi style" subset of [openbmb/Ultra-FineWeb-L3](https://huggingface.co/datasets/openbmb/Ultra-FineWeb-L3), using 2 epochs, 32 batch size, 1e-3 learning rate, and the [pika 4 tokenizer](https://huggingface.co/qikp/pika-4).

Also, evaluation on a tokenized and truncated [byunggill/gpt-2-output](https://huggingface.co/datasets/byunggill/gpt-2-output) was done during training.

## Limitations

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