bluejay

bluejay is an open-weights model that is built on a randomly initialized Gemma 3. I used the GPT-2 Tokenizer (from here) to tokenize the text, and the dataset used for training was pre-tokenized on another machine.

It was trained on Enron emails, using 1000 steps, 1 batch size, and 1 epoch.

Limitations

This model has a sub-optimal training pipeline, so it may output nonsensical responses.

Downloads last month
14
Safetensors
Model size
0.3B 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/bluejay-270m