FlashAttention<T>

Helpers & Utilities · AiDotNet.Tensors.Engines.Autodiff

No summary documentation available yet.

Methods

Method Summary
ValueTuple<Tensor<T>, Tensor<T>, Tensor<T>> Backward(Tensor<T> gradOutput, Tensor<T> query, Tensor<T> key, Tensor<T> value, Tensor<T> output, Tensor<T> logsumexp, int blockSizeQ, int blockSizeKV, double? scale, bool isCausal, int queryOffset, Tensor<T> attentionBias)
ValueTuple<Tensor<T>, Tensor<T>> Forward(Tensor<T> query, Tensor<T> key, Tensor<T> value, int blockSizeQ, int blockSizeKV, double? scale, bool isCausal, int queryOffset, Tensor<T> attentionBias)