Tags: pytorch/executorch
Tags
Arm backend: Replace module() calls with graph_module Using ExportedProgram.module() in later versions of torch introduce guards on the input. These guards cause an error in the ExportInterpreter which forbid the use of call_module. Fixes: #14417 Change-Id: I0cb68af9671dd1e3b543660c1cc49ff28cbffddc
Arm backend: Enable run on vulkan runtime by default and add xfails Change-Id: Iebf2f2a6a38f8540fafc3f43967e32d1261c89f8
Arm backend: Add docstrings for bmm and conv2d operators Change-Id: I603e09f79de4d87c0fdfe23d0a5bb87c14a5b61f Signed-off-by: Sebastian Larsson <[email protected]>
Arm Backend: Add support for bitwise_not.default Signed-off-by: Agrima Khare <[email protected]> Change-Id: Ie6fa82f61e40bbecea0ad31e3f6bcb305455ded2
Arm backend: Add special dtype TOSA handling Add a enum class to handle special dtypes that can't be represented in torch (i.e. int48_t) to avoid leaking serializer types into the pass handling of the backend. Signed-off-by: Per Åstrand <[email protected]> Change-Id: I3388cec3c8a26f28790eedc3f124c336b6724cb4
Fix _factory buck issue Signed-off-by: Erik Lundell <[email protected]> Change-Id: If014aa0fae6a0c16ace80a309b2bc490da69025c
Use appropraite setup scripts Change-Id: Ia0380fdd212c69728cdc52e720f3129f47a2d217
PreviousNext