According to your image
The pruning ratio is already 0.375% .
So, the parameters should be less in the pruned tlt model.
I am afraid you forget to set the pruned model in the retraining spec file.
For example,
model_config {
pretrained_model_file: “/workspace/tao-experiments/detectnet_v2/experiment_dir_pruned/resnet18_nopool_bn_detectnet_v2_pruned.tlt”
The size of the model file is reduced a lot, but “Trainable params” are the same in the training log, and the size of the retrained model is the same as the unpruned model.
Please upload your retraining spec file.
SPECS_retrain.txt (4.2 KB)
SPECS_train.txt (4.2 KB)
Please set the load_graph flag under model_config to true
See https://docs.nvidia.com/tao/tao-toolkit/text/object_detection/detectnet_v2.html#re-training-the-pruned-model or jupyter-notebook TAO Toolkit Quick Start Guide — TAO Toolkit 3.22.05 documentation