Quantized saved_model size increased.
JuHyung-Son opened this issue · 1 comments
JuHyung-Son commented
I followed colab tutorial and noticed that quantized mode size is increased after converting.
Is this an expected result?? I expected the quantized model size would be decreased.
I checked this PR 30789, but it doesn't work as expected.
TF 2.0.0
tensorrt 5.1.5
du -h -d 1
88K ./.config
448M ./resnet50_saved_model_TFTRT_FP16
386M ./resnet50_saved_model_TFTRT_INT8
924K ./data
298M ./resnet50_saved_model_TFTRT_FP32
103M ./resnet50_saved_model
55M ./sample_data
1.3G .
Tailwhip commented
I have same problem. I'm quantizing a model to reduce the size, not to make it larger :/