If transformer neural networks used 2x3 matrixes, then sure, we could use 'typical textbook illustration' to visualize them. The point of this tool is not to ex

If transformer neural networks used 2x3 matrixes, then sure, we could use 'typical textbook illustration' to visualize them. The point of this tool is not to explain matmul with toy examples, but to visualize real data from model weights.