feat & fix: support GQA/MQA and decode-phase attention via IAttentionLayer; add comprehensive HLO-level tests; fix bugs#4246

zewenli98 merged 5 commits into

mainpytorch/TensorRT:mainfrom

evanli/hlo-attention-testspytorch/TensorRT:evanli/hlo-attention-testsCopy head branch name to clipboard