ed by requantizing
on the fly, with a loss of accuracy. It is recommended to quantize the model with the
per-tensor scheme. See https://ai.google.dev/edge/litert/models/quantization_spec.