Merge pull request #6 from kpouget/fail

remoting: improve the failure when no virtgpu is available

Merge pull request #5 from kpouget/msg

remoting: improve the logging messages

Merge pull request #4 from kpouget/timeout

remoting: improve the loading timeout

graph : prepare for 4D mask (#14515)

ggml-ci

batch : add n_used count (#14512)

ggml-ci

CANN: Replace aclrtMemsetSync with aclnnInplaceZero operator (#14002)

Co-authored-by: luyuhong <luyuhong@kylinos.cn>

Merge pull request #2 from kpouget/dbg

remotingfrontend/virtgpu: give more time to load the libraries and log a warning on failure

ggml : implement GEGLU_ERF and GEGLU_QUICK ops (#14445)

opencl : broadcast for soft_max (#14510)

vulkan: support mixed/deepseekR1 FA head sizes (#14509)

* vulkan: better parameterize FA by head sizes

* vulkan: support mixed/deepseekR1 FA head sizes