Export trained model to onnx and Latency of network

[John1231983]

Thanks for the great library for hardware-aware NAS. I trained ProxylessNas with the CIFAR dataset. Here are the steps:

Step 1:

python main.py experiment=classification/mobilenet/cifar10_search_latency

Step 2:

python main.py experiment=classification/mobilenet/cifar10_train_latency

I have three questions:

1. After training the model (step 2), how can I export the trained model to ONNX?
2. Regarding latency, how can I set a target latency so that the latency of the trained model is smaller than the expected latency?
3. The latency used on the repo that is on the host device, how can I perform a search with latency constraints based on the latency observed on the target/edge device?