feat(tabby): Add experimental vulkan support

[boxbeam]

No description provided.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 50.33%. Comparing base (7f90102) to head (affcb48).
Report is 7 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1588      +/-   ##
==========================================
+ Coverage   49.98%   50.33%   +0.34%     
==========================================
  Files         112      112              
  Lines        8994     9068      +74     
==========================================
+ Hits         4496     4564      +68     
- Misses       4498     4504       +6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wsxiaoys]

related:
#631
#124

[amne]

in src/engine.cc:365 the llama_backend_init(...) function call is missing the mandatory bool argument.
This is breaking the build for me and I don't understand how it passed?

cargo build --features cuda

  cargo:warning=src/engine.cc: In constructor ‘llama::{anonymous}::BackendInitializer::BackendInitializer()’:

  cargo:warning=src/engine.cc:365:23: error: too few arguments to function ‘void llama_backend_init(bool)’

  cargo:warning=  365 |     llama_backend_init();

  cargo:warning=      |     ~~~~~~~~~~~~~~~~~~^~

  cargo:warning=In file included from src/engine.cc:11:

  cargo:warning=llama.cpp/llama.h:267:20: note: declared here

  cargo:warning=  267 |     LLAMA_API void llama_backend_init(bool numa);

  cargo:warning=      |                    ^~~~~~~~~~~~~~~~~~

  exit status: 0
  exit status: 1

[boxbeam]

@amne I'm wondering if your llama.cpp submodule is up to date. If you cloned Tabby a while ago, you need to run git submodule update --init --recursive. The required parameter was removed in the version we are now using.