Taehee Jeong comments

Results 15 comments of


                                            Taehee Jeong

Structural (filter) pruning for convolutional layers

Thanks for your interest in contribution! Please read [contribution instructions](https://github.com/tensorflow/model-optimization/blob/master/CONTRIBUTING.md) to take further steps. As this looks like a whole new feature, you also might want to [file an RFC](https://github.com/tensorflow/model-optimization/blob/master/CONTRIBUTING_TECHNIQUE.md)

quantization.keras.quantize_model function runtime error with mobilenet v3

Hi @ejcgt, can you confirm if this issue still happens on latest version?

Does anyone try to apply this project to BERT?

Hi @haozha111, is there some updates?

Support for Conv1D Quantization

Hi @jennakwon06, can you update?

Support `float16` input for `tf.quantization.fake_quant_*` ops

Hi @Xhark , do you think it should be handled by MOT team or TF core team?

Tensorflow1.15 and TensorFlow2.4 were used for int8 precision quantization of the same Pb model, and the quantization results were different. Where, the BIAS of float32 bit is the same, but the weight of int8 bit convolution kernel is different

Not sure about different first layer in 1.15 vs 2.4. For weight difference - is the left result from TF 2.4? Recent TFLite supports per-channel quantization for conv-like ops, so...

Taehee Jeong

Structural (filter) pruning for convolutional layers

quantization.keras.quantize_model function runtime error with mobilenet v3

Does anyone try to apply this project to BERT?

Support for Conv1D Quantization

Support `float16` input for `tf.quantization.fake_quant_*` ops

Tensorflow1.15 and TensorFlow2.4 were used for int8 precision quantization of the same Pb model, and the quantization results were different. Where, the BIAS of float32 bit is the same, but the weight of int8 bit convolution kernel is different

Is it possible to have quantised fused_tanh (with add op ) via TFlite converter

problem with quantizing the BN layer

QAT (quantization aware training) Support quantizing models recursively

Pruning: Keras subclassed model increased support