Script 'mail_helper' called by obssrc Hello community, here is the log from the commit of package llamacpp for openSUSE:Factory checked in at 2026-07-13 14:25:49 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Comparing /work/SRC/openSUSE:Factory/llamacpp (Old) and /work/SRC/openSUSE:Factory/.llamacpp.new.1991 (New) ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "llamacpp" Mon Jul 13 14:25:49 2026 rev:31 rq:1365268 version:9964 Changes: -------- --- /work/SRC/openSUSE:Factory/llamacpp/llamacpp.changes 2026-06-27 18:09:33.246990481 +0200 +++ /work/SRC/openSUSE:Factory/.llamacpp.new.1991/llamacpp.changes 2026-07-13 14:26:07.263425182 +0200 @@ -1,0 +2,43 @@ +Sun Jul 12 11:16:07 UTC 2026 - Markus Ebner <[email protected]> + +- Enable OpenVINO Backend + +------------------------------------------------------------------- +Sat Jul 11 14:08:29 UTC 2026 - Eyad Issa <[email protected]> + +- Update to version 9964: + * Server: timings/progress on /responses API, fixed deadlock in + load_models(), enforced prompt cache RAM limit, improved tools + handling (removed apply_diff), fixed draft model fit vs load. + * CUDA: fused MMVQ post-scale for NVFP4, K-type validation for + flash attention, removed -sm row, refactored cuBLAS, concat for + quantized types. + * Vulkan: disabled FA mask_opt on GCN, fixed 32-bit overflow in + CEIL_DIV, reduced submission threshold for small AMD GPUs, + added src0 type checks. + * OpenCL: cluster-parallel decode FA for Adreno, ragged-tile MoE + prefill FP16 GEMM, general FA decode optimizations, Q6_K GEMM/ + GEMV fix for non-multiple-of-128 weights. + * SYCL: added col2im_1d, cross_entropy_loss ops, fixed UT cases + for CONT, CPY, noncontiguous ACC, enhanced argsort. + * Metal: added CONV_2D_DW, set_rows with f16 src0, col2im_1d op. + * WebGPU: tuned subgroup split (d_split) in flash_attn_vec. + * Hexagon: added VISION RoPE, new VTCM layouts, improved + pipelines for MUL_MAT, MUL_MAT_ID, FLASH_ATTN_EXT, + tiling/tracing. + * UI: context usage gauge/panel, pre-select models via alias, + server modalities in non-router mode, Ctrl+B sidebar toggle, + improved streaming, export full message tree. + * llama/llama-batch: KQ masks f16 if FA used, fixed decreasing + pos, added unit tests, added n_keep_tail in split_equal. + * ggml: smaller chunks in CUDA ggml_top_k/argsort, fixed conv 2d + dw, added CPU f16->f16 SET_ROWS support. + * CLI: added --output option, HTTP-based implementation, fixed + crash on wrong server base url. + * Speculative decoding: fixed ngram-map OOB read on prompt + shrink, fixed stale tensor-split params for draft models. + * Security: fixed OOB reads in UGM tokenizer. + * Full commit log: + https://github.com/ggml-org/llama.cpp/compare/b9760...b9964 + +------------------------------------------------------------------- Old: ---- llama-b9760-ui.tar.gz llamacpp-9760.tar.gz New: ---- llama-b9964-ui.tar.gz llamacpp-9964.tar.gz ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Other differences: ------------------ ++++++ llamacpp.spec ++++++ --- /var/tmp/diff_new_pack.xq2zcf/_old 2026-07-13 14:26:09.087487548 +0200 +++ /var/tmp/diff_new_pack.xq2zcf/_new 2026-07-13 14:26:09.095487822 +0200 @@ -25,11 +25,11 @@ %global mtmd_sover 0.0.%{version} %global mtmd_sover_suffix 0 -%global ggml_sover 0.15.2 +%global ggml_sover 0.16.0 %global ggml_sover_suffix 0 Name: llamacpp -Version: 9760 +Version: 9964 Release: 0 Summary: Inference of Meta's LLaMA model (and others) in pure C/C++ License: MIT @@ -44,6 +44,8 @@ BuildRequires: shaderc BuildRequires: spirv-headers BuildRequires: pkgconfig(OpenCL) +BuildRequires: pkgconfig(OpenCL-CLHPP) +BuildRequires: pkgconfig(openvino) BuildRequires: pkgconfig(libcurl) BuildRequires: pkgconfig(vulkan) # 32bit seems not to be supported anymore @@ -132,6 +134,15 @@ This package includes the OpenCL backend for ggml. +%package -n libggml-openvino +Summary: A tensor library for C++ (OpenVINO backend) + +%description -n libggml-openvino +A tensor library for C++. It was created originally to support llama.cpp +and WhisperCpp projects. + +This package includes the OpenVINO backend for ggml. + %package -n ggml-devel Summary: Development files for ggml Obsoletes: libggml < 7266 @@ -192,6 +203,7 @@ -DGGML_CPU=ON \ -DGGML_VULKAN=ON \ -DGGML_OPENCL=ON \ + -DGGML_OPENVINO=ON \ -DGGML_BACKEND_DL=ON \ -DGGML_BACKEND_DIR="%{backend_dir}" \ -DGGML_OPENCL_USE_ADRENO_KERNELS=OFF \ @@ -266,6 +278,11 @@ %dir %{backend_dir} %{backend_dir}/libggml-opencl.so +%files -n libggml-openvino +%license LICENSE +%dir %{backend_dir} +%{backend_dir}/libggml-openvino.so + %files -n ggml-devel %license LICENSE %{_includedir}/ggml*.h ++++++ llama-b9760-ui.tar.gz -> llama-b9964-ui.tar.gz ++++++ ++++ 7556 lines of diff (skipped) ++++++ llamacpp-9760.tar.gz -> llamacpp-9964.tar.gz ++++++ /work/SRC/openSUSE:Factory/llamacpp/llamacpp-9760.tar.gz /work/SRC/openSUSE:Factory/.llamacpp.new.1991/llamacpp-9964.tar.gz differ: char 16, line 1
