11. Support

Post questions about TI TVM to the TI E2E™ design community forum and select the TI device being used.

11.1. Unsupported Layer Configs on Hardware NPU Accelerator

When the compilation runs into an unsupported layer configuration on the hardware NPU accelerator, the compiler outputs a warning message and possible fixes. For example:

WARNING:ti_mcu_nnc:GCONV layer NOT offloaded to TI-NPU
WARNING:ti_mcu_nnc:GCONV layer config: 8-bit input [93, 1, 14], 8-bit output [45, 1, 14], 8-bit kernel [4, 1], strides [2, 1], padding [(0, 0), (0, 0)], data layout NHWC, kernel layout HWIO
WARNING:ti_mcu_nnc:GCONV does not support 14 input channels and 14 output channels. Please change to multiples of 4.

WARNING:ti_mcu_nnc:FCONV layer NOT offloaded to TI-NPU
WARNING:ti_mcu_nnc:FCONV layer config: 8-bit input [256, 1, 1], 8-bit output [128, 1, 8], 8-bit kernel [5, 1], strides [2, 1], padding [(2, 2), (0, 0)], data layout NHWC, kernel layout HWIO
WARNING:ti_mcu_nnc:Current FCONV implementation only supports certain layer configs. Please follow TI Neural Network Compiler for MCUs User's Guide to check supported layer configs and request support for your layer config.

Please include these warning messages when asking for support.