How xLA and TensorFlow Worked Together

The approach was basically to rip out the numeric side of a core language that are existed, the type system everything rip that out kind of replace it. And by the way, you put in hooks for a compiler to kind of work back and forth. Yeah, it was very, like I said, it was a little fragile early on so I think when I first started I was still like a very young immature programmer that depended on stack overflow for a lot of my work. At the time there's really no like I couldn't just Google how to build xLA for another language. A lot of things that you just kind of had to figure out on your ownIt was almost like the Wild