Time series data with multiple features from TFRecords to keras lstm model

Question

I am trying to predict one of around 80 classes based on a TFRecords file with about 50 features. Some of the features are int64 and some are float. One TFRecords file has about