oa::FnMatrix::scaledDotProductAttention

scaledDotProductAttention: shared semantic SDPA operation. The standard provider materializes probabilities; the explicit Flash compatibility API uses an online causal provider. An empty mask means no additive mask.

Function Documentation

Matrix oa::FnMatrix::scaledDotProductAttention( const Matrix & inQ, const Matrix & inK, const Matrix & inV, const Matrix & inAdditiveMask, oa::F32 inScale, bool inCausal = false )
scaledDotProductAttention: shared semantic SDPA operation. The standard provider materializes probabilities; the explicit Flash compatibility API uses an online causal provider. An empty mask means no additive mask.

Parameters

inQ
const Matrix &

inK
const Matrix &

inV
const Matrix &

inAdditiveMask
const Matrix &

inScale
oa::F32

inCausal
bool

Default: false

Returns

Matrix

The declared return value.