I’ve been testing the MPT models and they’re better than all the other open source stuff I’ve tried. I’m using my task-based, one-shot prompts that are about 1-

I’ve been testing the MPT models and they’re better than all the other open source stuff I’ve tried. I’m using my task-based, one-shot prompts that are about 1-2K tokens. Seems pretty on par with GPT-3.5. Marketing or whatever, first time I’ve been at all impressed.