jsGPT Lab

Experiment with jsGPT, an implementation of NanoGPT written in TensorflowJS!

Tune the hyperparameters below to your liking and hit "train" to train our NanoGPT model on Shakespeare's The Tragedy of Coriolanus. Over time, you should see a fairly significant improvement in output quality, as the model learns to mimic the Shakespearean writing style.

Note: please use a browser that supports WebGPU, WebGL, or WASM. This model runs on your own machine, NOT a backend — don't go too crazy with the hyperparameters.

Inputs

Architectural Hyperparameters

Parameters that influence the model's internal architecture

Context Size:

# of Embeddings:

# of Transformers:

# of Heads:

Head Size:

Dropout:

Training Hyperparameters

Parameters that influence the model's training

Batch Size:

Learning Rate:

Eval Interval:

Max Iterations:

Outputs

Confused about the hyperparameters?

Want to try a simpler model?

What is jsGPT?