mul_mat_id

This release introduces a group matmul direct API for mul_mat_id in ggml-zendnn and scales the MUL_MAT_ID fallback threshold by expert count. It also provides numerous pre-compiled binaries for various platforms and hardware configurations.

Introduces the initial ET backend for ggml, significantly expanding the supported operations and optimizations. This release includes numerous bug fixes, performance enhancements, and new features for various kernels and backends.

This release updates the OpenVINO backend to version 2026.2.1, introduces self-contained release packages, and includes several operator improvements for the OpenVINO backend.

This release introduces NVFP4 support within the Vulkan backend for several core tensor operations. It also provides a comprehensive set of pre-compiled binaries for diverse operating systems and hardware accelerators.

This release introduces significant enhancements to the Hexagon backend by adding support for IQ4_NL and MXFP4 quantization types, alongside fixes for mixed-quant models and code formatting.

This release focuses on bug fixes, primarily addressing issues within the Vulkan backend related to quantization overflow and matrix multiplication failures across different hardware configurations.

This release optimizes Vulkan performance by preprocessing expert IDs to more efficiently manage workgroups. It also provides updated binaries for a wide range of operating systems and hardware backends including CUDA 12/13, SYCL, and HIP.

This release focuses on performance optimizations for the Vulkan backend, specifically targeting the coopmat2 mul_mat_id shader by improving memory access patterns and reducing overhead.

This release optimizes the Vulkan backend by adjusting the block size (BK=32) for cooperative matrix multiplication. It includes a wide range of pre-built binaries for multiple architectures and acceleration backends.