QAT on Image classification -- faulty documantation?

Please provide the following information when requesting support.

• Hardware (RTX3090)
• Network Type (Classification)
• TLT Version (latest)

The TAO image classification documentation states in the notes section the following: “When exporting a model trained with Quantization Aware Training (QAT) enabled, the tensor scale factors to calibrate the activations are peeled out of the model and serialized to a TensorRT readable cache file defined by the cal_cache_file argument.”

While there’s nothing in the specfiles and adding enable_qat: true gives an error. Is it possible to use QAT in (re)training an image classifier?

There is no enable_qat parameter for classification training. We will modify the document.
Currently, only post training quantization is supported.

There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.