Add support for llama-cpp GPU acceleration

[moejay]

Add n_gpu_layers param to llama-cpp that allows some processing on the GPU

This is in preparation for added GPU support in llama-cpp
abetlen/llama-cpp-python#203 for when is approved and merged

This is the llama-cpp commit that this PR adds support for

Before submitting

• No New integration

Who can review?

• @hwchase17
• @agola11

Thanks all for the wonderful project!

[m0sh1x2]

Is there an easy way to test all of the functionality in one go or there is quite a lot of chained merges below?

[moejay]

Is there an easy way to test all of the functionality in one go or there is quite a lot of chained merges below?

not that I can think of (This is dependent on one PR though) so maybe it's not too bad to test

you need the llama-cpp-python binding to be updated, and for that
you could pip install from https://github.com/moejay/llama-cpp-python (That's my branch with this updated)

I can add some more detailed testing instructions in the PR description when I come back later this evening if that helps.

I did notice that a bunch of the examples were broken (There are probably issues for those that exist at the moment, maybe I'll be able to contribute some fixes there later on)

[moejay]

Looks like the same change was merged a bit later here
So, this should now work with the latest llama-cpp-bindings

[dev2049]

duplicate of #4739