Dear QC Engineer,
We're using QNN to load a 3B machine learning model. However, we got the following memory error while doing so. Can you kindly assist me on how to debug & fix the issue? Is this a inherent limitation of QNN or Hexagon DSP?
We're loading the model from a binary context to avoid composing the graph on device (S23 Snapdragon Gen 2 SM8550).[+2]:[DEBUG] Calling QNN function: contextCreateFromBinary [+0]:[ERROR] <E> fastrpc memory map for fd: 45 with length: 450887680 failed with error: 1 [+0]:[ERROR] <E> Failed to map weights buffer to device! [+0]:[ERROR] <E> Could not allocate persistent weights buffer! [+0]:[ERROR] <E> Failed to initialize graph memory [+0]:[ERROR] <E> Failed to initialize graph with id 257 context 4 deviceId 0 coreId 0 pdId 0 with err 1002 [+0]:[ERROR] <E> Context create from binary failed for deviceId 0 coreId 0 pdId 0 err 1002
RPC Memory error from QNN
Posted: Fri, 2023-09-29 04:17