[Bug Report] Review current matmul function usages

[bryce13950]

Describe the bug

Originally from @ArthurConmy via slack

We seem to use batch_addmm stuff (link) as GPT-2 uses Conv1D (link), but this doesn't really fix the problem as Pythia (link), Llama (link) etc do not use Conv1D and I think their matmul implementations are not batchaddmm at all

We probably need to review all usages for various models to make sure we have the right usages that are needed for specific models. As the project has grown, the implementations have been moved around a lot, and we need to figure out a way to route through a larger variety of components based on config to handle a larger variety of requirements.

System Info
Describe the characteristic of your environment:

• Describe how transformer_lens was installed (pip, docker, source, ...)
• What OS are you using? (Linux, MacOS, Windows)
• Python version (We suppourt 3.7 -3.10 currently)

Additional context
Add any other context about the problem here.

Checklist

• I have checked that there is no similar issue in the repo (required)