hi , i have question regarding the BYOC modules .
lets the BYOC module is getting executed on a custom hardware/accelerator which is using some part of GPU memory reserved for custom hardware.
According to my observation right now , the BYOC module’s output is automatically stored in HOST,
inspite of the calculation being in accelerator (GPU kinda) device ,
and i would like to know can we force tvm to not explicitly copy the BYOC module output to host so that we can use this output as input to other gpu/byoc modules.?
many thanks