Does executorch support qat quant for qualcomm qnn backend?

[Novelfor]

I want to use qat method for my model, but i can only find ptq quantizer in executorch, are there some examples of how to implement Quantization Aware Training (QAT) for qnn backend?

[jerryzh168]

QAT is going through the same code path as PTQ. if you don't have some special handling for conv-bn, you could start with adding the configuration for QAT, like: https://github.com/pytorch/pytorch/blob/a0429c01ad665ffb2faa04a411913ecee9962566/torch/ao/quantization/quantizer/xnnpack_quantizer.py#L116

[jerryzh168]

@cccclai can you tag qualcomm POC to see if they want to support QAT?

[cccclai]

Yeah QAT is supported. cc: @shewu-quic @chunit-quic @haowhsu-quic @winskuo-quic let's add some docs