torch.backends.quantized.supported_engines
Out[3]: [‘qnnpack’, ‘none’]
fbgemm
is not supported.
I tried to install fbgemm
on my Ubuntu 20 and reinstalled pytorch 1.8. The problem is still there.
and also I tried to use qnnpack
to run Quantization Aware Training on Bert. No matter I set the Embedding layer to be ‘None’ or ‘float16_dynamic_qconfig’, I got a error message:
Didn’t find engine for operation quantized::linear_prepack NoQEngine