reduce memory load with quantization

noticed that we are using 32 bit precision for at least some of the model components (maybe the data too?). I was reading about quantization (reducing precision) to make neural net params smaller in memory. https://www.tensorflow.org/performance/quantization

maybe worth considering?

No files or tests are named; start by locating the model components and data representations in the repository, then check where 32-bit precision is used. Assess whether quantization applies to those components and measure memory usage; done means an implemented approach that reduces the model's memory load.

Written by the indexing model from the issue text.