How to train on multiple possible outputs to the same question efficiently in an LLM?
How would you try an LLM with multiple different outputs to the same question efficiently without recalculating the context each time? For example:
How would you try an LLM with multiple different outputs to the same question efficiently without recalculating the context each time? For example: