BVLC/caffe ? reverse-engineered prompt

Build me a fast deep learning framework for vision that feels like Caffe, with clear modular pieces, good speed, and simple examples I can actually run.

I want the project to include the core library, command line tools, Python support, and a few example models so someone can train and test a network without too much setup. Please make the installation and build steps straightforward, and include a working sample workflow for image classification or feature extraction. If there are any parts that depend on current build docs or platform quirks, look up the latest info online and make the setup as smooth as you can.

The main thing is that a new user should be able to clone it, build it, and follow a small tutorial to see it working end to end. Keep the structure clean, document the key steps, and make sure the examples and default models are easy to find.