AWS realtime predictions using boto3

I am very confused how to get realtime predictions with the predict() command in boto3. First of all every other prediction and model creation you simply use csv files, why is this different, to reduce latency? and is there any way to use an s3 file as that would be preferable even if it would increase latency. The second thing that confuses me is the documentation seemingly states that value part of the dictionary must be a string. Many of my values are not strings (boolean or ints) and I still need realtime predictions. I think that saying the value has to be a string may not be accurate and may simply just be unintentionally misleading though. I don't have a realtime endpoint setup yet so I have not tested to see if it would through an exception. I just wanted to do some research before I set everything up. Any help would be greatly appreciated!

So basically: 1. Is there any way to use a csv file with realtime predictions 2. If not, how do I go about formatting the record section of the api call so that I can use things other than strings.