A newer version of this model is available: qikp/kite-4.1-14m

Kite

🎉 You are looking at Kite 3.5, which is now using the Qwen3 architecture, a different dataset and an efficient configuration!

Kite is a small, trained, 10 million parameter language model.

Training

It was trained on 50K rows of a tokenized version of Cosmopedia v0.1 using 1 epoch, 32 batch size, 1.5e-4 learning rate, and the pika 3 tokenizer.

Limitations

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

Downloads last month
82
Safetensors
Model size
9.38M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train qikp/kite-3.5-10m