Deep Learning with Tensorflow - Quantization Aware Training

#tensorflow #machinelearning #deeplearning

Quantization aware training (QAT) allows for reduced precision representations of weight. In QAT model is quantized as part of training process itself as against post model quantization

quantization aware training is often better for model accuracy than post quantization training.

quantization allows inference to be carried out using integer-only arithmetic, which can be implemented more efficiently than floating point inference on commonly available integer-only hardware

[jashsingh]

can we use this for conv1D as well? all the threads that i have read tells me that it is not yet supported.

[pawanreddyulindala]

Hello sir, How can I access this colab file or Github link of the ipynb file

[suewhoo]

why my quantized model is bigger than the non-quantized model..?

[vijayakakumani]

Nice explaination on Quantization! How to incorporate batch normalization layer, as its giving me error: Layer batch_normalization_7: is not supported. You can quantize this layer by passing a instance to the `quantize_annotate_layer` API

[tahamansoor]

ipynb or git link for notebook???'

[vignesha]

Hi, I am not of CS background, what is the reason behind putting data in cache

[AmanKumarSharma-deft]

Nice video ...really informative ..though the glitches in the sound can be improved ....can we used this in tensorflow 1.14 version?

[Janamejaya.Channegowda]

Great content. Please try to avoid speaking when you run a cell. Audio gets messy at that time.