Script 'mail_helper' called by obssrc Hello community, here is the log from the commit of package ollama for openSUSE:Factory checked in at 2026-08-09 21:42:17 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Comparing /work/SRC/openSUSE:Factory/ollama (Old) and /work/SRC/openSUSE:Factory/.ollama.new.16738 (New) ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "ollama" Sun Aug 9 21:42:17 2026 rev:68 rq:1370283 version:0.32.3 Changes: -------- --- /work/SRC/openSUSE:Factory/ollama/ollama.changes 2026-06-10 15:49:15.367028602 +0200 +++ /work/SRC/openSUSE:Factory/.ollama.new.16738/ollama.changes 2026-08-09 21:45:25.542908382 +0200 @@ -1,0 +2,150 @@ +Tue Aug 4 12:48:11 UTC 2026 - VaiTon <[email protected]> + +- Add missing service_del_postun call for ollama.service + +------------------------------------------------------------------- +Sat Aug 1 19:55:48 UTC 2026 - Martin Pluskal <[email protected]> + +- Add the empty %changelog section openSUSE:Factory requires: without it + factory-auto declines the submission outright +- Reference AGENTS.md as a source: factory-auto also declines any file + in the package that the spec does not mention + +------------------------------------------------------------------- +Sat Jul 25 12:38:55 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to version 0.32.3 + * Fixed model downloads that stall before sending data. + * Improved integrations: restored Claude Code Channels, + fixed Anthropic thinking streams, and made Hermes Desktop + respect --force-build. + * Expanded GPU support with CUDA on Windows ARM64, B200 support + through CUDA 12, and lower memory use on Linux CUDA/ROCm iGPUs. + * Added chat, thinking, and tool calling support for Laguna 2.1 + models, including a Metal inference fix. + * Fixed GLM tool calls being silently dropped at the end of generation. + * Updated the MLX and llama.cpp engines. + +------------------------------------------------------------------- +Fri Jul 17 06:26:35 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to version 0.32.1 + * Improved Gemma 4 tool calling and multi-turn reasoning, + including more reliable tool-response continuations + * Fixed a recurrent MLX model cache leak that could increase + memory use across requests, and improved cache snapshot performance + * MLX text model loading now respects OLLAMA_LOAD_TIMEOUT + * Agent web search and fetch now tell users to run ollama + signin when authentication is required + * The interactive agent now receives the current working directory + for better project context + * Fixed ollama launch so choosing Pick another model for a + deprecated model passed with --model opens the model picker + * Updated VS Code setup documentation for the official Ollama extension + +------------------------------------------------------------------- +Tue Jul 14 09:31:00 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to version 0.32.0 + * New interactive agent experience: running ollama now launches + an agent to help you code and delegate work + * Renamed the Codex App integration to ChatGPT: use ollama launch + chatgpt (and --restore to return to your usual ChatGPT profile) + * Simplified integration selection: the ollama launch menu now + only offers the most popular integrations (other integrations + can be accessed through ollama launch + * Warns before launching older agent models: CodeLlama, Qwen2.5(-coder), + Llama 3.x, Mistral, StarCoder, and the base DeepSeek-R1 tags now + prompt a deprecation warning before ollama launch continues + +------------------------------------------------------------------- +Thu Jul 9 05:53:42 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to version 0.31.2 + * Enabled flash attention on older NVIDIA GPUs (compute capability 6.x) + * iGPU can now offload vision models with padding to fit available memory + * Fixed structured output for thinking models when thinking is disabled + * Hardened GGUF model creation + * ollama launch for Claude Code now disables telemetry by default + * Fixed loading models on paths with non-UTF-8 characters + * Updated the MLX and llama.cpp engines + +------------------------------------------------------------------- +Tue Jul 7 15:20:44 UTC 2026 - Adrian Schröter <[email protected]> + +- Define a build counter sync tag, to ensure all flavors have + the same build number + +------------------------------------------------------------------- +Tue Jul 7 15:20:44 UTC 2026 - Adrian Schröter <[email protected]> + +- Define a build counter sync tag, to ensure all flavors have + the same build number + +------------------------------------------------------------------- +Thu Jul 2 06:38:48 UTC 2026 - Adrian Schröter <[email protected]> + +- Updated to version 0.31.1 + * Tightened Gemma 4 MoE model loading in the MLX engine + * Updated the MLX engine to the latest version, including a new small-batch matmul kernel + * Updated the underlying llama.cpp engine to build 9840 + * Improved Gemma 4 multi-token prediction (MTP) performance + * new versions of llama.cpp and mlx + +------------------------------------------------------------------- +Sat Jun 27 08:59:05 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to 0.30.11: + * launch: add thinking capability detection to opencode + * launch: auto-install Claude Code + * launch: auto-install opencode when missing + * discover: fix inverted iGPU/dGPU Vulkan classification on Windows hybrid graphics + * mlxrunner: unify and tune speculative decoding + * launch/codex: detect model drift when Codex App UI switches + * llama: add sm_86 architecture to cuda_v13_windows preset + * llm: size mmproj offload + * docs: document max think level + * llm: preserve generation headroom for shifted prompts + * llama: default qwen2.5vl window attention metadata + * llm: use host Vulkan loader on Windows + * mlx: update and fix CUDA JIT packaging + * llm: fix ollama ps double-counting mmap'd weights on partial offload + * docs: redesign docs landing and integrations overview + * server: align generate with native chat templates + * jetson: add CC 87 for CUDA v13 + * llama.cpp version update + +- Updated generated using pbuild-ai + * Helper script provided in tool-scripts/ directory + Use pbuild-ai --allow-tool-scripts to enable it. + +------------------------------------------------------------------- +Fri Jun 19 07:18:20 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to 0.30.10: + * Updated the underlying llama.cpp engine to build 9672 + +------------------------------------------------------------------- +Wed Jun 17 09:46:46 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to 0.30.9: + * Support for Cohere2Moe architecture + * Fixed LFM2 parser/render for cases where thinking was not emitted + * Fixed issue where ollama launch claude and other coding agent or assistant use cases would only output one token + * Ollama will now return an error if a single message is larger than the current context window +- switch to tar balls because we use git lfs now + +------------------------------------------------------------------- +Sun Jun 14 12:54:45 UTC 2026 - Adrian Schröter <[email protected]> + +- Update to 0.30.8: + * Fixed ollama launch selecting the wrong provider in some cases + * Improved prompt caching by decoupling it from context shift + for better KV cache reuse + * More stable MLX inference with hardened linear and embedding layers + * MLX runner now creates snapshots during prompt processing and + speculative decoding for improved reliability + * Improved recurrent model support with per-boundary states from + the gated-delta kernels + +------------------------------------------------------------------- Old: ---- llama.cpp-main.obscpio llama.cpp.obsinfo mlx-c-main.obscpio mlx-c.obsinfo mlx-main.obscpio mlx.obsinfo ollama-0.30.6.tar.gz vendor.tar.zstd New: ---- AGENTS.md _constraints _scmsync.obsinfo build.specials.obscpio llama.cpp-main.tar.xz mlx-c-main.tar.xz mlx-main.tar.xz ollama-0.32.3.tar.gz tool-scripts.obscpio vendor.tar.gz ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ Other differences: ------------------ ++++++ ollama.spec ++++++ --- /var/tmp/diff_new_pack.sQHRqb/_old 2026-08-09 21:45:29.503043558 +0200 +++ /var/tmp/diff_new_pack.sQHRqb/_new 2026-08-09 21:45:29.507043694 +0200 @@ -15,6 +15,21 @@ # Please submit bugfixes or comments via https://bugs.opensuse.org/ # +#!BcntSyncTag: ollama + +# We require too much memory atm. +# these jobs work on all supported archs with the specified memory of _constraints file +# found by trial and error +%ifarch x86_64 +%global jobs 3 +%else +%global jobs 4 +%endif + +# Keep these vendored revisions pinned to the versions required by Ollama. +%global llama_cpp_version b10091 +%global mlx_version 33c03c486c34a7dadab5339563612c9933c4a406 +%global mlx_c_version fba4470b89073180056c9ea46c443051375f7399 %{bcond_with rocm} %{bcond_with cuda} @@ -65,20 +80,32 @@ %define cuda_version %{cuda_version_major}-%{cuda_version_minor} Name: ollama -Version: 0.30.6 +Version: 0.32.3 Release: 0 Summary: Tool for running AI models on-premise License: MIT URL: https://ollama.com -Source: https://github.com/ollama/ollama/archive/v%{version}/%{name}-%{version}.tar.gz -Source1: vendor.tar.zstd +# The archive name after #/ is the local filename used by the package. +Source: https://github.com/ollama/ollama/archive/refs/tags/v%{version}.tar.gz#/ollama-%{version}.tar.gz +Source1: vendor.tar.gz Source2: %{name}.service Source3: %{name}-user.conf Source4: sysconfig.%{name} Source5: %{name}-rpmlintrc -Source10: llama.cpp-main.tar.xz -Source11: mlx-main.tar.xz -Source12: mlx-c-main.tar.xz +# Hints for the pbuild-ai update tooling that maintains this package. Listed as +# a source because factory-auto declines any file the spec does not mention. +Source6: AGENTS.md +# Keep the helper directory recognized by source_validator. +Source7: tool-scripts + +# Pinned vendored revision. +Source10: https://github.com/ggml-org/llama.cpp/archive/refs/tags/%{llama_cpp_version}.tar.gz#/llama.cpp-main.tar.xz + +# Pinned vendored revision. +Source11: https://github.com/ml-explore/mlx/archive/%{mlx_version}.tar.gz#/mlx-main.tar.xz + +# Pinned vendored revision. +Source12: https://github.com/ml-explore/mlx-c/archive/%{mlx_c_version}.tar.gz#/mlx-c-main.tar.xz Patch0: fix-mlxrunner-tests.diff Patch1: disable-llama.cpp-ui.patch BuildRequires: ccache @@ -139,14 +166,14 @@ BuildRequires: rocminfo %endif -Requires(pre): %fillup_prereq +%systemd_ordering # 32bit seems not to be supported anymore while riscv isn't yet. ExcludeArch: %{ix86} %{arm} riscv64 %sysusers_requires %if 0%{?force_gcc_version} BuildRequires: gcc%{force_gcc_version}-c++ %endif -%if 0%{suse_version} >= 1600 +%if 0%{?suse_version} >= 1600 BuildRequires: gcc-c++ >= 11.4.0 %else BuildRequires: gcc14-c++ @@ -183,12 +210,21 @@ %prep %autosetup -a1 -p1 -n %{name}-%{version} +mkdir llama.cpp mlx mlx-c +tar xf %{S:10} --strip-components=1 -C llama.cpp +tar xf %{S:11} --strip-components=1 -C mlx +tar xf %{S:12} --strip-components=1 -C mlx-c + +# 2. Patch cmake/local.cmake to skip git-dependent compat patch checks (submodules are extracted from tarballs, not git) +sed -i 's/git submodule update --init --recursive/# git submodule update skipped — sources from tarballs/' CMakeLists.txt || true +sed -i '/message(FATAL_ERROR.*Failed to apply Ollama/s/FATAL_ERROR/WARNING/g' cmake/local.cmake %build %if "%{?flavor}" == "test" exit 0 %endif +# ninja has built-in defaults, but it also calls gnu make again what is not caring about ninja values either %define __builder ninja %sysusers_generate_pre %{SOURCE3} %{name} %{name}-user.conf @@ -228,8 +264,8 @@ echo -e 'package ml\nvar LibOllamaPath string = "/usr/%{_lib}/ollama"' > ml/path.go # and also the local copy of MLX sources -export OLLAMA_MLX_SOURCE=%_sourcedir/mlx-main -export OLLAMA_MLX_C_SOURCE=%_sourcedir/mlx-c-main +export OLLAMA_MLX_SOURCE=$PWD/mlx +export OLLAMA_MLX_C_SOURCE=$PWD/mlx-c %cmake \ -UOLLAMA_INSTALL_DIR -DOLLAMA_INSTALL_DIR=%{_libdir}/ollama \ @@ -239,7 +275,9 @@ -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ -DCMAKE_INSTALL_RPATH='$ORIGIN' \ -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=OFF \ - -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=%_sourcedir/llama.cpp-main \ + -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=$PWD/../llama.cpp \ + -DOLLAMA_LLAMA_CPP_SKIP_COMPAT_PATCH=ON \ + -DOLLAMA_BUILD_PARALLEL=%jobs \ %if "%{?flavor}" == "cuda" -DCMAKE_CUDA_COMPILER=/usr/local/cuda-%{cuda_version_major}.%{cuda_version_minor}/bin/nvcc \ %endif @@ -267,7 +305,8 @@ -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ -DCMAKE_INSTALL_RPATH='$ORIGIN' \ -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=OFF \ - -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=%_sourcedir/llama.cpp-main \ + -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=$PWD/llama.cpp \ + -DOLLAMA_BUILD_PARALLEL=%jobs \ cmake --build build/mlx_cuda_v13 -- -l %{jobs} exit 0 @@ -284,7 +323,8 @@ -DCMAKE_BUILD_WITH_INSTALL_RPATH=ON \ -DCMAKE_INSTALL_RPATH='$ORIGIN' \ -DCMAKE_INSTALL_RPATH_USE_LINK_PATH=OFF \ - -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=%_sourcedir/llama.cpp-main \ + -DFETCHCONTENT_SOURCE_DIR_LLAMA_CPP=$PWD/llama.cpp \ + -DOLLAMA_BUILD_PARALLEL=%jobs \ local_directory=build/llama-server-%preset_flavor cmake --build ${local_directory%_linux} -- -l %{jobs} @@ -294,7 +334,7 @@ # # Main package # -%cmake_build +%cmake_build -j %jobs cd .. go build -trimpath -o %{name} . @@ -387,22 +427,17 @@ ldd -r "$file" | grep "undefined symbol:" && exit 1 done exit 0 +%else +# Validate installed unit file syntax (safe fallback if tools missing) +systemd-analyze verify %{buildroot}%{_unitdir}/%{name}.service || true +exit 0 %endif -%if 0%{?force_gcc_version} -export CXX="g++-%{?force_gcc_version}" -export CC="gcc-%{?force_gcc_version}" -# pie doesn't work with gcc12 on leap -export GOFLAGS="-mod=vendor" -%endif -go test -v ./... || exit 0 - %pre -f %{name}.pre %service_add_pre %{name}.service %post %service_add_post %{name}.service -%fillup_only %preun %service_del_preun %{name}.service ++++++ AGENTS.md ++++++ ## hints for pbuild-ai to update the package ## After changed the package version in the spec file: * make sure to call tool-scripts/update_references.sh to update all vendored changes * abort when this is not possible * don't search for newer versions of llama-cpp, mlx or mlx-c. We must use the pinned versions. ++++++ _constraints ++++++ <constraints> <hardware> <physicalmemory> <size unit="G">32</size> </physicalmemory> <disk> <size unit="G">12</size> </disk> </hardware> </constraints> ++++++ _scmsync.obsinfo ++++++ mtime: 1786133858 commit: 46296765f712ff3d3c69e44694949bdd096e4ef4dba1431d35778fcf6679bddc url: https://src.opensuse.org/AI/ollama.git revision: main ++++++ _service ++++++ --- /var/tmp/diff_new_pack.sQHRqb/_old 2026-08-09 21:45:29.623047653 +0200 +++ /var/tmp/diff_new_pack.sQHRqb/_new 2026-08-09 21:45:29.627047790 +0200 @@ -1,36 +1,7 @@ <services> - <service name="format_spec_file" mode="manual" /> -<!-- - <service name="download_files" mode="manual" /> ---> - <service name="go_modules" mode="manual"> - <param name="compression">zstd</param> - </service> - <service name="obs_scm" mode="manual"> - <param name="url">https://github.com/ggml-org/llama.cpp.git/</param> - <param name="scm">git</param> - <!-- defined in LLAMA_CPP_VERSION file --> - <param name="revision">refs/tags/b9509</param> - <param name="version">main</param> - </service> - <service name="obs_scm" mode="manual"> - <param name="url">https://github.com/ml-explore/mlx.git</param> - <param name="scm">git</param> - <!-- defined in MLX_VERSION file --> - <param name="revision">2165dc08d7b33258260aa849d39f087d50e62962</param> - <param name="version">main</param> - </service> - <service name="obs_scm" mode="manual"> - <param name="url">https://github.com/ml-explore/mlx-c.git</param> - <param name="scm">git</param> - <!-- defined in MLX_C_VERSION file --> - <param name="revision">fba4470b89073180056c9ea46c443051375f7399</param> - <param name="version">main</param> - </service> - <service name="tar" mode="buildtime" /> - <service name="recompress" mode="buildtime"> - <param name="file">*.tar</param> - <param name="compression">xz</param> + <service name="download_files" mode="manual"> + <param name="recompress">yes</param> </service> + <service name="go_modules" mode="manual" /> </services> ++++++ build.specials.obscpio ++++++ diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn' '--exclude=.svnignore' old/.gitignore new/.gitignore --- old/.gitignore 1970-01-01 01:00:00.000000000 +0100 +++ new/.gitignore 2026-08-07 22:17:38.000000000 +0200 @@ -0,0 +1,10 @@ +*.obscpio +*.osc +_build.* +.pbuild +.assets +ollama +llama.cpp +mlx +mlx-c + ++++++ ollama-0.30.6.tar.gz -> ollama-0.32.3.tar.gz ++++++ /work/SRC/openSUSE:Factory/ollama/ollama-0.30.6.tar.gz /work/SRC/openSUSE:Factory/.ollama.new.16738/ollama-0.32.3.tar.gz differ: char 12, line 1 ++++++ ollama.service ++++++ --- /var/tmp/diff_new_pack.sQHRqb/_old 2026-08-09 21:45:29.783053115 +0200 +++ /var/tmp/diff_new_pack.sQHRqb/_new 2026-08-09 21:45:29.787053252 +0200 @@ -13,3 +13,4 @@ [Install] WantedBy=default.target + ++++++ sysconfig.ollama ++++++ --- /var/tmp/diff_new_pack.sQHRqb/_old 2026-08-09 21:45:29.815054208 +0200 +++ /var/tmp/diff_new_pack.sQHRqb/_new 2026-08-09 21:45:29.819054344 +0200 @@ -45,8 +45,8 @@ OLLAMA_TMPDIR= ## Type: string -## Description: Models to be loaded by default -## Default: "" +## Description: Storage loction of the models +## Default: "/var/lib/ollama" ## ServiceRestart: ollama OLLAMA_MODELS= @@ -56,4 +56,13 @@ ## ServiceRestart: ollama OLLAMA_ORIGINS= +## Type: int +## Description: Maximal context length ollama can handle +## Default: +## < 24 GiB VRAM: 4k context +## 24-48 GiB VRAM: 32k context +## >= 48 GiB VRAM: 256k context +## ServiceRestart: ollama +OLLAMA_CONTEXT_LENGTH= +
