- Published on
Multiple figures exploring patterns within the feed forward network weight matrices.
Latest
Some writings on my interests and work.
- Published on
Multiple figures exploring how the model's prediction of the next token evolves through the layers of the model.- Published on
Interesting plots examining properties of the model parameters in the attention block.- Published on
Interesting plots showing how the transformer attention mechanism behaves.- Published on
An overview of the setup I use for the research. Which model, dataset, ect...