Automate upstream llama.cpp sync

[bgs4free]

I'm trying to build with a newer upstream version of llama.cpp. Syncing the changes from upstream llama.h into llama_cpp.py is quite a toil. Has anyone (successfully) attempted to use LLM assistance to make it less painful?

[m-bappe-therman7777]

If you are able to build, could you share the llama_cpp.py (updated) and the version of llama.cpp you're using? Thanks in advance!

[bgs4free]

If you are able to build, could you share the llama_cpp.py (updated) and the version of llama.cpp you're using? Thanks in advance!

Unfortunately I'm not. Currently attempting to get afa8a9e of the upstream project running. There are lots of breaking changes I'd have to understand first and which will affect more than just llama_cpp.py. Some functions seem gone and I'm not sure how to replace them, if at all.

I have no prior experience with this project nor llama.cpp. Just gave it a shot because I got impatient waiting for the deepseek tokenizer.

[JamePeng]

Maybe you can try my repaired version. #1901

[bgs4free]

Maybe you can try my repaired version. #1901

Yes, can confirm this works. Can use Deepseek R1 distills now. Great work!