Migrating to TAO 5.5#

The major change in this release is the deprecation of command line parameters in place of experiment specfile fields. The only flag that will now be accepted is -e for the experiment specfile. All other arguments must correspond to a field in the experiment configuration.

The table below shows examples of these updates to the CLI. This is not an exhaustive list of all actions, but these changes can be generalized to all networks included as part of TAO 5.5.

Details on which fields are available to configure and more examples can be found in each respective network’s documentation.

Network

TAO 5.2.x

TAO 5.5.x

PyTorch

 
tao model <network> train -e $SPECS_DIR/train.yaml \
                          -r $RESULTS_DIR \
                          -k $KEY \
                          --gpus 2 \
                          train.num_epochs=2

tao model <network> train -e $SPECS_DIR/train.yaml \
                          results_dir=$RESULTS_DIR \
                          encryption_key=$KEY \
                          train.num_gpus=2 \
                          train.num_epochs=2

TF2

 
tao model <network> train -e $SPECS_DIR/train.yaml \
                          --gpus 1 \
                          --gpu_index 0 \
                          train.num_epochs=2

tao model <network> train -e $SPECS_DIR/train.yaml \
                          num_gpus=1 \
                          gpu_ids=[0] \
                          train.num_epochs=2

Data-Services

 
tao dataset augmentation generate -e $SPECS_DIR/augment.yaml \
                                  -r $RESULTS_DIR \
                                  --gpus 2 \
                                  data.image_dir=$IMAGE_DIR

tao dataset augmentation generate -e $SPECS_DIR/augment.yaml \
                                  results_dir=$RESULTS_DIR \
                                  num_gpus=2 \
                                  data.image_dir=$IMAGE_DIR

Deploy

 
tao deploy <network> gen_trt_engine -e $SPECS_DIR/gen_trt_engine.yaml \
                                    -r $RESULTS_DIR \
                                    -k $KEY \
                                    --gpu_index 1 \
                                    gen_trt_engine.onnx_file=$ONNX_FILE
                                    gen_trt_engine.trt_engine=$ENGINE_PATH

tao deploy <network> gen_trt_engine -e $SPECS_DIR/gen_trt_engine.yaml \
                          results_dir=$RESULTS_DIR \
                          encryption_key=$KEY \
                          gen_trt_engine.gpu_id=1 \
                          gen_trt_engine.onnx_file=$ONNX_FILE
                          gen_trt_engine.trt_engine=$ENGINE_PATH