Hi, I'm working with the TinyML Hello World project to perform static inference on a model. When I set CONV_DEPTHW_MODE to STANDARD and enable the accelerator, I indeed notice a faster inference speed. However, the output generated by the model changes significantly compared to when the accelerator is not enabled, resulting in a substantial decrease in accuracy. Could you please advise on what might be causing this issue? Does your convolution accelerator utilize any approximate algorithms?
Hi, I'm working with the TinyML Hello World project to perform static inference on a model. When I set CONV_DEPTHW_MODE to STANDARD and enable the accelerator, I indeed notice a faster inference speed. However, the output generated by the model changes significantly compared to when the accelerator is not enabled, resulting in a substantial decrease in accuracy. Could you please advise on what might be causing this issue? Does your convolution accelerator utilize any approximate algorithms?