4 ms·
When prototyping or developing new model code, yes. If you’re building production ML systems, there’s plenty code to write and test before deploying to prod in
by roseway4 6y ago
When prototyping or developing new model code, yes. If you’re building production ML systems, there’s plenty code to write and test before deploying to prod infrastructure.
To be clear, you would work with a very small data sample or synthetic data as your objective isn’t to train a model for production use.
Edit: clarity and grammar
- skwb 6y agoYeah, exactly this. I mostly am using cloud VMs for training, but sometimes just need to be able to mess around on my local machine for my deployment pipeline. I don't really care if it takes ~10 seconds to do inference on a handful of images, I really just need to tinker with the interfaces to ensure everything is hooked up correctly.