It doesn't matter if you have the RAM, running a 1.5TB model for a single context stream is fundamentally inefficient.