fix: add cuda llava native libraries.

[AsakusaRinne]

The files are from Action 8518332362, with no updating for llama.cpp commit.

[martindevans]

One of the files is called LLama\runtimes\deps\cu11.7.1\libllama_shared.so, should that be LLama\runtimes\deps\cu11.7.1\libllava_shared.so (llava instead of llama)?

[AsakusaRinne]

One of the files is called LLama\runtimes\deps\cu11.7.1\libllama_shared.so, should that be LLama\runtimes\deps\cu11.7.1\libllava_shared.so (llava instead of llama)?

Yes, I misspelled that in the workflow. I have fixed this typo and manually renamed the file before submitting this PR.

[AsakusaRinne]

@martindevans I think we should publish a patch release since it's a severe performance issue, as shown in #639 and #650. Do you have other works that want to be included together in 0.11.2?

[martindevans]

Shall we merge #649 and then do a patch release with this PR?

We'll probably need to do another patch release in a few days to fix the publish issue we were discussing on Discord. That's not really a problem though.

[AsakusaRinne]

Shall we merge #649 and then do a patch release with this PR?

I think it's a good idea.

We'll probably need to do another patch release in a few days to fix the publish issue we were discussing on Discord. That's not really a problem though.

That's why it's called patch release! :)

[martindevans]

@AsakusaRinne #649 has been reviewed and merged, you can go ahead and merge this one when you're ready :)