A server built on top of a long-running Caffe process with a fixed architecture.
Make sure caffe/python is on your PYTHONPATH, otherwise barista will be confused.
export PYTHONPATH=$PYTHONPATH:<path-to-caffe/python>
If you already have an initialized model, you can start the server with
python -m barista <architecture.prototxt> <model.caffemodel>
Otherwise, you'll need to specify a solver which can create a new model instance. For example:
python -m barista models/deepq/train_val.prototxt models/deepq/deepq.caffemodel --solver models/deepq/solver.prototxt
This will create a new instance of the model architecture defined in train_val.prototxt, save it to deepq.caffemodel and fire up the barista server with this new model instance.