llama.cpp Adds Vulkan-based Tensor Parallel Support

An update is underway in the llama.cpp project to improve Tensor Parallel (TP) functionality so it can be used practically even in environments using the Vulkan backend.

This work aims to optimize model parallel processing in distributed computing environments via Vulkan, improving inference efficiency on hardware that uses this API.

Key points are as follows:

• Usability improvements for the Vulkan Tensor Parallel feature
• Optimization of inference performance in distributed environments
• Strengthened support for related hardware acceleration