Ask HN: Are you running production workloads in Pytorch eager mode?

I have an idea for optimizations using a dedicated backend for pytorch. It only applies to eager mode execution. From my understanding, eager mode is used for development and debugging. Are there real production workloads that work in that way ? I assume people optimize by compiling graphs using JIT but then again, people are lazy...