https://www.orionsarm.com/eg-article/51519aa6bab33

1. You simply download the file and keep it. We’re talking about stuff you can run on your own computer, not services like chatgpt or such.

2. Highly variable. Depends on model size and format. For LLMs, range would be from under 1GB up to a current maximum of ~240GB.

3. It’s a big file of numbers (“weights” or “parameters”) either integers or floating point depending on format.