# llama.cpp adds TOP_K_MASK to Vulkan backend

> The open‑source llama.cpp library updates its Vulkan support to include a TOP_K_MASK feature, fixing related barriers and header declarations.

Oossa · 2026-11-13 · https://oossa.com/en/llama-cpp-adds-top-k-mask-to-vulkan-backend

The ggml‑org/llama.cpp project released a new version on 2026‑11‑13. The update adds a TOP_K_MASK operation to the Vulkan GPU backend, fixes a barrier issue for that mask, and declares the function in the common header. The changes were made with assistance from Claude Opus 5.5.

## The facts

- Release date: 2026‑11‑13
- Vulkan backend now supports TOP_K_MASK

## Why it matters

GPU users can now apply top‑k masking during inference, giving finer control over token selection.

## Sources & references

1. [ggml-org/llama.cpp b12365](https://github.com/ggml-org/llama.cpp/releases/tag/b12365) – llama.cpp, 2026-11-13

Last updated: 2026-11-13