github llama.cpp · 2h ago · inference

llama.cpp b10442 adds Vulkan SHMEM_STRIDE_PAD for Intel Xe and fixes OOB read

from English

The llama.cpp project released version b10442, introducing specific Vulkan optimizations for Intel Xe hardware while addressing memory safety issues in matrix multiplication routines.

Adds SHMEM_STRIDE_PAD and APPLY_SLM_A_RESHAPE flags for coopmat mul_mm on Intel Xe.
Fixes shared memory estimation for Intel SHMEM_STRIDE_PAD=0 in matmul_shmem_support.
Implements cacheline alignment for shared kvalues_mxfp4 to improve data handling.
Resolves an out-of-bounds (OOB) read in kvalues_mxfp4 initialization following the cacheline padding changes.
Restricts SLM-A reshape adjustments to Intel Windows drivers and reverts mxfp4 cacheline padding where necessary.

This release provides updated binaries for macOS, Linux, Windows, Android, and openEuler across CPU, Vulkan, CUDA, ROCm, OpenVINO, and SYCL backends.

Importance 1/3 Trust 1/3 llama.cpp Inference efficiency Local inference

Read original

github llama.cpp · 22h ago · 2 views

llama.cpp b10434 passes reasoning_effort to templates

The llama.cpp project released build b10434, which adds support for passing the `reasoning_effort` parameter through to Jinja templates. This change ensures that OpenAI Chat Completions `reasoning_effort` values are stored and made available during generation.

github llama.cpp · 2d ago

llama.cpp b10434 passes reasoning_effort to templates

llama.cpp b10418 adds SYCL host pinned memory support

llama.cpp OpenVINO backend adds Qwen3.5 support and memory optimizations

llama.cpp b10414 adds TQ2_0 support and optimizes mul_mv kernel

llama.cpp b10354 fixes CPU affinity mask on Android