While there has been a lot of progress in leveraging GPT models for various use-cases, the model weights themselves are closed and only accessible via API. As an internal team within a startup, we don't have the resources to train our own model. We have loads of proprietary/internal/client data that we'd like to use but are concerned about sending it to third-party APIs like OpenAI's.
Are there any comparisons available between the various models with publicly available weights vs ChatGPT? How should we go about choosing the model to use? Is it the case that beyond a certain size of the model, say 10B parameters, there is very little difference in performance between two models? Additionally, are there any examples of people building tools for internal usage within a startup/enterprise? Is it possible to run GPT models locally for inference and fine-tune them for our specific use-case? What are the steps required to achieve this, and what are some best practices to follow?
There’s already GPT4All that can run locally in your computer with just CPU.
Pretty sure by this time next week there will be several companies or projects built around this in different flavors: self-hosted, cloud, hybrid, distributed…
Also an open plug-ins architecture protocol so that plug-ins can work across multiple different LLMs/models/services