Script 'mail_helper' called by obssrc Hello community, here is the log from the commit of package llamacpp for openSUSE:Factory checked in at 2026-07-31 15:27:19 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Comparing /work/SRC/openSUSE:Factory/llamacpp (Old) and /work/SRC/openSUSE:Factory/.llamacpp.new.2004 (New) ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "llamacpp" Fri Jul 31 15:27:19 2026 rev:32 rq:1368396 version:10078 Changes: -------- --- /work/SRC/openSUSE:Factory/llamacpp/llamacpp.changes 2026-07-13 14:26:07.263425182 +0200 +++ /work/SRC/openSUSE:Factory/.llamacpp.new.2004/llamacpp.changes 2026-07-31 16:09:00.467978333 +0200 @@ -1,0 +2,55 @@ +Wed Jul 22 08:06:55 UTC 2026 - Andreas Schwab <[email protected]> + +- Enable OpenVINO backend only on x86_64 and aarch64 + +------------------------------------------------------------------- +Tue Jul 21 23:49:41 UTC 2026 - Markus Ebner <[email protected]> + +- Update to version 10078 + * GGML: bump to version 0.17.0 + * Vulkan: add support for Q2_0 quantization + * CUDA: vectorize same-type get_rows with int4 copy + * OpenCL: optimize Q4_K/Q5_K flat matrix-vector kernels by loading quants as uint + * OpenCL: improve Q4_K performance with transposed no-shuffle scales for coalesced reads + * OpenCL: optimize MoE DP4A activation tile access using 128-bit local memory transfers + * OpenCL: add broadcast support for Adreno MUL_MAT and honor view_offs + * OpenCL: add ABS operation + * DFlash: rotate injected K/V cache + * DFlash/Eagle3: automatically download Hugging Face sidecar files + * DeepSeek V4: write only used rows to model state + * Quantization: exclude DeepSeek V4 FFN routing table (ffn_gate_tid2eid) from quantization + * Vulkan: refactor queue handling to use per-instance mutexes and unique handles + * OpenVINO: improve dynamic backend loading support + * Hexagon: add CLAMP operation + * UI: add conversation bulk actions and improve Settings/Sidebar + * UI: enable agentic flow when only the JavaScript sandbox is active + * UI: fix Settings → Display tool call visibility toggle + * UI: fix collapsed user bubble markdown rendering + * Full commit log: + https://github.com/ggml-org/llama.cpp/compare/b10054...b10078 + +------------------------------------------------------------------- +Fri Jul 17 01:59:30 UTC 2026 - Markus Ebner <[email protected]> + +- Update to version 10054: + * SYCL: add fused top-k MoE + * Vulkan: Use native e2m1 and e4m3 conversions for mxfp4/nvfp4 + * SYCL: set fattn_vec_nthreads to 256 for Battlemage + * Vulkan/CPU: Support f16 as SET_ROWS src + * SYCL: Flash Attention with XMX engine via oneDNN + * SYCL: implement xielu op + * SYCL: support kernel type fp16 for conv2d_dw + * DeepseekV4: reduce graph splits + * ui: fix thinking menu never appearing in single-model mode + * CUDA: dedup MoE gate/up activation quantization + * Enable CUDA graphs on volta+turing + * CUDA: Support CUDA Virtual Devices + * Full commit log: + https://github.com/ggml-org/llama.cpp/compare/b9964...b10054 + +------------------------------------------------------------------- +Fri Jul 17 01:53:30 UTC 2026 - Markus Ebner <[email protected]> + +- Enable multiversioned build for the CPU backend + +------------------------------------------------------------------- Old: ---- llama-b9964-ui.tar.gz llamacpp-9964.tar.gz New: ---- llama-b10078-ui.tar.gz llamacpp-10078.tar.gz ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Other differences: ------------------ ++++++ llamacpp.spec ++++++ --- /var/tmp/diff_new_pack.zYlVeF/_old 2026-07-31 16:09:01.059998928 +0200 +++ /var/tmp/diff_new_pack.zYlVeF/_new 2026-07-31 16:09:01.059998928 +0200 @@ -25,11 +25,17 @@ %global mtmd_sover 0.0.%{version} %global mtmd_sover_suffix 0 -%global ggml_sover 0.16.0 +%global ggml_sover 0.17.0 %global ggml_sover_suffix 0 +%ifarch x86_64 aarch64 +%bcond_without openvino +%else +%bcond_with openvino +%endif + Name: llamacpp -Version: 9964 +Version: 10078 Release: 0 Summary: Inference of Meta's LLaMA model (and others) in pure C/C++ License: MIT @@ -45,8 +51,10 @@ BuildRequires: spirv-headers BuildRequires: pkgconfig(OpenCL) BuildRequires: pkgconfig(OpenCL-CLHPP) -BuildRequires: pkgconfig(openvino) BuildRequires: pkgconfig(libcurl) +%if %{with openvino} +BuildRequires: pkgconfig(openvino) +%endif BuildRequires: pkgconfig(vulkan) # 32bit seems not to be supported anymore ExcludeArch: %{ix86} %{arm} @@ -201,9 +209,12 @@ -DLLAMA_CURL=ON \ -DGGML_NATIVE=OFF \ -DGGML_CPU=ON \ + -DGGML_CPU_ALL_VARIANTS=ON \ -DGGML_VULKAN=ON \ -DGGML_OPENCL=ON \ +%if %{with openvino} -DGGML_OPENVINO=ON \ +%endif -DGGML_BACKEND_DL=ON \ -DGGML_BACKEND_DIR="%{backend_dir}" \ -DGGML_OPENCL_USE_ADRENO_KERNELS=OFF \ @@ -266,7 +277,7 @@ %files -n libggml-cpu %license LICENSE %dir %{backend_dir} -%{backend_dir}/libggml-cpu.so +%{backend_dir}/libggml-cpu-*.so %files -n libggml-vulkan %license LICENSE @@ -278,10 +289,12 @@ %dir %{backend_dir} %{backend_dir}/libggml-opencl.so +%if %{with openvino} %files -n libggml-openvino %license LICENSE %dir %{backend_dir} %{backend_dir}/libggml-openvino.so +%endif %files -n ggml-devel %license LICENSE ++++++ llama-b9964-ui.tar.gz -> llama-b10078-ui.tar.gz ++++++ ++++ 7440 lines of diff (skipped) ++++++ llamacpp-9964.tar.gz -> llamacpp-10078.tar.gz ++++++ /work/SRC/openSUSE:Factory/llamacpp/llamacpp-9964.tar.gz /work/SRC/openSUSE:Factory/.llamacpp.new.2004/llamacpp-10078.tar.gz differ: char 16, line 1
