1 Choose a model
not loaded
2 Ask something
3 Explore the run
After you generate, every forward pass shows up here: the tokens the model was given, the scores it gave every possible next token, and the one it picked.