sycl

This release optimizes SYCL quantized KV decode for BMG hardware by routing it to TILE, while keeping VEC for other architectures until further validation.

This release marks tq2_0 as not supported for SYCL and provides updated pre-built binaries for various platforms including macOS, Linux, Android, and Windows with different acceleration options.

This release re-adds SYCL Q2_K reordered MMVQ and ESIMD kernels. It also includes various pre-compiled binaries for different operating systems and hardware configurations.

This release bumps the llama.cpp version to 0.2.0 and ggml to 0.21.0, incorporating numerous backend improvements, bug fixes, and new model support across various hardware accelerators like SYCL, OpenCL, Metal, and Vulkan.

Optimized SYCL quantized cpy kernel launches by adjusting thread/block counts for better resource utilization. This change significantly boosts performance on specific quantization paths.

Optimized SYCL backend by fusing operations for dense FFN layers, leading to measurable performance gains on specific hardware. New pre-built binaries are available for various platforms and configurations.

This release introduces performance improvements for SYCL by fusing the gated-delta-net state writeback copy operation. It also provides updated pre-compiled binaries for various platforms and hardware accelerators.

SYCL GEMM has been updated to use automatic FP16 promotion, simplifying type handling. This release also provides pre-compiled binaries for various platforms and hardware accelerators.

This release addresses a bug in SYCL GPU code related to UE4M3 parsing, ensuring correct handling of unsigned scale values. Various pre-compiled binaries for different platforms and hardware accelerators are provided.

This release includes a fix for a SYCL-related error on arc770 hardware. It also provides updated pre-compiled binaries for various platforms and configurations.

This release includes a fix for SYCL iGPU classification and provides updated pre-compiled binaries for various platforms and backends.

This release introduces SYCL support for dev2dev memcpy, enhancing performance for specific hardware configurations. It also provides updated pre-compiled binaries for various platforms and configurations.

This release includes performance optimizations for the SYCL backend by fusing RMS_NORM and MUL operations. It also provides updated pre-compiled binaries for various platforms and hardware configurations.

This release enhances the SYCL backend with performance improvements for unary elementwise operations, including a contiguous fast path and 32-bit index math. It also introduces optimizations for elementwise index calculations using fast division.

This release includes several bug fixes for the SYCL backend, addressing issues with row calculations, kernel processing, and boundary checks. These fixes aim to improve the stability and correctness of SYCL operations.

This release includes a fix for SYCL quantization types and provides pre-compiled binaries for various platforms and hardware configurations.

This release introduces the implementation of the xielu operation for the SYCL backend. It also provides pre-compiled binaries for various platforms and hardware configurations, including macOS, Linux, Android, and Windows.

This release includes SYCL backend optimizations for Battlemage hardware and provides a comprehensive set of pre-built binaries for a wide range of platforms and hardware configurations.

This release introduces fused top-k MoE for the SYCL backend, enhancing performance. It also provides updated pre-compiled binaries for various platforms and hardware configurations.

This release introduces SYCL backend improvements with Q2_K support in DMMV reorder path. It also provides pre-compiled binaries for various platforms including macOS, Linux, Android, and Windows with support for multiple backends like Vulkan, ROCm, OpenVINO, SYCL, CUDA, and HIP.

This release addresses an AOT double type issue related to SYCL functionality and provides a comprehensive set of pre-built binaries across macOS, Linux, Android, Windows, and UI targets.

This release updates environment variable naming conventions for SYCL features, changing 'disable' prefixes to 'enable' prefixes, and provides numerous pre-compiled binaries for various platforms.

This release primarily addresses bug fixes within the SYCL implementation, resolving failed unit tests. It also provides updated binary distributions for numerous operating systems and hardware configurations.

This release includes a fix for the SYCL reorder function and enhances the build script by adding explicit support for SYCL FP32/FP16 compilation targets. Numerous platform binaries are also provided.

This release introduces an optimization for SYCL by routing small f32 matmuls to oneMKL. It also provides extensive pre-compiled binaries across multiple operating systems and hardware configurations.

This release introduces performance improvements to the SYCL backend by reordering MMVQ/dequant paths for Q5_K and Q8_0 quantization, alongside removing duplicate definitions.

This release enhances the performance of the fattn operation specifically for the SYCL backend. It also provides a comprehensive set of pre-built binaries for numerous operating systems and hardware configurations.

This release addresses a bug in the SYCL implementation by fixing an incorrect variable check via assert and migrating to the GGML API for that component.

This release addresses a bug related to the untransposed GDA recurrent state when using the SYCL backend. It also provides updated binary distributions for numerous operating systems and hardware configurations.

This release introduces BF16 conversion support for Intel oneAPI via SYCL and announces a change in Linux distribution formats from .zip to .tar.gz.