Source/Packages

Realization.Nvidia.SM86.ExactCrossEntropyBackwardSM86

packages/realizations/cooperative/nvidia-sm86/src/Realization/Nvidia/SM86/ExactCrossEntropyBackwardSM86.alpha

702 lines118 declarations31.2 KiBSHA-256 adcd1f5d6195

family · lines 19–24

ExactCrossEntropyBackwardSM86FailureCode

Full file
Exact full-vocabulary logit gradient for coppelius. The preceding exact forward pass supplies loss[r] = logsumexp(logits[r]) - correct[r], so this kernel reconstructs logsumexp as loss + correct and writes (exp(logit - logsumexp) - one_hot(target)) / rows. Grid Y selects one of 48 vocabulary tiles, avoiding a source-level 48-way instruction expansion.
19family ExactCrossEntropyBackwardSM86FailureCode : Type 0
20constructor ExactCrossEntropyBackwardInstructionCountMismatch
21constructor ExactCrossEntropyBackwardEncodedByteCountMismatch
22constructor ExactCrossEntropyBackwardEncodingFailed
23constructor ExactCrossEntropyBackwardIdentityFailed
24constructor ExactCrossEntropyBackwardIdentityLengthInvalid

The compiler supplied declaration spans and resolved links from this source snapshot. This page does not assert that this file belongs to a checked closure.