Make MultiHeadAttention op return attention probabilities#23125

Open

amancini-N wants to merge 2 commits intomicrosoft:mainfrom amancini-N:attn-probs-mha

+321-66

Commits on Dec 11, 2024

Allow returning attention probs from MultiHeadAttention
amancini-N
committed

Commits on Dec 12, 2024

Add CUDA implementation for attn_probs
amancini-N
committed