Bob's training fixture places one example for each current-byte row. The
dedicated one-block realization walks each complete 256-logit row, emits
exact full-vocabulary loss and gradient, and performs no host work.
12def bobLossGradientImage =
13 bobBigramLossGradientImageThe compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.