3 ms·
I'm not sure open source applies to actual models. Models aren't human readable, so it's closer to a binary blob. It would apply to the training code and possib
by api 3y ago
I'm not sure open source applies to actual models. Models aren't human readable, so it's closer to a binary blob. It would apply to the training code and possibly data set.
Llama2 is a binary blob pre-trained model that is useful and is licensed in a fairly permissive way, and that's fine.
- politelemon 3y agoYes I think you've put it well. If models were smaller I'd see those in the Github releases section. The model training is what I'd see in the source code and the README etc, to arrive at the 'blob'.
- api 3y agoEven if it costs millions in compute to run at that scale, seeing that code would be extremely informative.
- cjdell 3y agoVery like a binary blob. You have to execute it to use it and impossible for humans to reason about just by looking at it. At least binary blobs can be disassembled.