InteractiveVisualizations
GloVe
Inside the Model
Explore the co-occurrence matrix, see how well the model reconstructs log(X), compare the two sets of vectors W and W̃, and trace how biases absorb word frequency.
Inside GloVe digs into the model with real trained data across 200 epochs. You can inspect the co-occurrence matrix, evaluate factorization quality via reconstruction error, compare W and W̃ vectors, and explore bias-frequency correlation and PMI approximation. Training convergence is tracked per word pair across all epochs.