[CUDA] Serve quantized KV caches with XQA: attention sinks, independent and per-channel scales - #29900

[CUDA] Serve quantized KV caches with XQA: attention sinks, independent and per-channel scales#29900

Tianlei Wu (tianleiwu) merged 5 commits into

mainmicrosoft/onnxruntime:mainfrom

tlwu/20260725/increase_xqa_coveragemicrosoft/onnxruntime:tlwu/20260725/increase_xqa_coverageCopy head branch name to clipboard