onnxruntime
Upgrade cudnn_frontend to 1.24 and enable cuDNN SDPA for MHA/GQA
#28849
Merged

Commits
  • cudnn frontend 1.24 for MHA and GQA
    tianleiwu committed 136 days ago
  • Fix CUDA plugin Windows link failure by linking NVRTC
    Copilot committed 136 days ago
  • address feedbacks
    tianleiwu committed 134 days ago
  • address review feedback: reuse computed members for cuDNN flash flags
    tianleiwu committed 134 days ago
  • Fix MHA cuDNN SDPA auto path to honor is_stable(); drop stale GQA comment
    tianleiwu committed 134 days ago
Loading