This is a tool to help you understand the inner workings of the Othello-GPT model, specifically the ever elusive MLP layers!
Made by Albert Garde and Esben Kran.
The table below shows the 2,048 neurons in each of the 8 layers of the Othello-GPT Transformer model. The row number indicates the "rank" that is decided by sorting according to the neuron's activation variance as a proxy for interest.
Click on a neuron to see the examples that activate it.