One pass through the model, top to bottom. Every shape and count is computed from the configuration above rather than drawn. Open a stage to branch out the arithmetic inside it.
visualise the transformer, one stage at a time
One pass through the model, top to bottom. Every shape and count is computed from the configuration above rather than drawn. Open a stage to branch out the arithmetic inside it.
hover the map to read a weight
same prompt, same seed, one architecture at a time. switch the model to see what the four Phase 2 changes are actually worth.
press generate