It uses a redis cache? I wonder if future version could use a global cache somewhere on the internet, and use cloud solvers. Seems like you need a redis cache

It uses a redis cache? I wonder if future version could use a global cache somewhere on the internet, and use cloud solvers. Seems like you need a redis cache per version, but that shouldn't be hard to do -- or just add a version key to each cache request. Seems weird to have this type of tool as a local tool rather than a cloud based tool. It would be neat to have a service you can query with a snippet of IR, and it gives you back an optimal version of it. One can continually increase the side of these snippets as you get more cloud resources. Having statistics on which patterns are common, even if they are not optimized would be really interesting. You could then use these statistics to target future optimization research.