Merge pull request #300567 from abysssol/update-ollama-0.1.30

ollama: 0.1.29 -> 0.1.30

pkgs/tools/misc/ollama/cmake-include.patch

@@ -1,7 +0,0 @@
---- a/llm/llama.cpp/examples/server/CMakeLists.txt
-+++ b/llm/llama.cpp/examples/server/CMakeLists.txt
-@@ -11,3 +11,4 @@
-     TARGET_LINK_LIBRARIES(${TARGET} PRIVATE ws2_32)
- endif()
- target_compile_features(${TARGET} PRIVATE cxx_std_11)
-+include (../../../ext_server/CMakeLists.txt) # ollama

pkgs/tools/misc/ollama/default.nix

@@ -28,14 +28,32 @@
 
 let
   pname = "ollama";
-  version = "0.1.29";
+  # don't forget to invalidate all hashes each update
+  version = "0.1.30";
+
   src = fetchFromGitHub {
     owner = "jmorganca";
     repo = "ollama";
     rev = "v${version}";
-    hash = "sha256-M2G53DJF/22ZVCAb4jGjyErKO6q2argehHSV7AEef6w=";
+    hash = "sha256-+cdYT5NUf00Rx0fpCvWUNg4gi+PAOmZVDUdB3omibm0=";
     fetchSubmodules = true;
   };
+  vendorHash = "sha256-Lj7CBvS51RqF63c01cOCgY7BCQeCKGu794qzb/S80C0=";
+  # ollama's patches of llama.cpp's example server
+  # `ollama/llm/generate/gen_common.sh` -> "apply temporary patches until fix is upstream"
+  # each update, these patches should be synchronized with the contents of `ollama/llm/patches/`
+  llamacppPatches = [
+    (preparePatch "03-load_exception.diff" "sha256-1DfNahFYYxqlx4E4pwMKQpL+XR0bibYnDFGt6dCL4TM=")
+    (preparePatch "04-locale.diff" "sha256-r5nHiP6yN/rQObRu2FZIPBKpKP9yByyZ6sSI2SKj6Do=")
+  ];
+
+  preparePatch = patch: hash: fetchpatch {
+    url = "file://${src}/llm/patches/${patch}";
+    inherit hash;
+    stripLen = 1;
+    extraPrefix = "llm/llama.cpp/";
+  };
+
 
   validAccel = lib.assertOneOf "ollama.acceleration" acceleration [ null "rocm" "cuda" ];
 
@@ -45,6 +63,7 @@ let
   enableRocm = validAccel && (acceleration == "rocm") && (warnIfNotLinux "rocm");
   enableCuda = validAccel && (acceleration == "cuda") && (warnIfNotLinux "cuda");
 
+
   rocmClang = linkFarm "rocm-clang" {
     llvm = rocmPackages.llvm.clang;
   };
@@ -94,12 +113,6 @@ let
       buildGo122Module.override { stdenv = overrideCC stdenv gcc12; }
     else
       buildGo122Module;
-  preparePatch = patch: hash: fetchpatch {
-    url = "file://${src}/llm/patches/${patch}";
-    inherit hash;
-    stripLen = 1;
-    extraPrefix = "llm/llama.cpp/";
-  };
   inherit (lib) licenses platforms maintainers;
 in
 goBuild ((lib.optionalAttrs enableRocm {
@@ -110,8 +123,7 @@ goBuild ((lib.optionalAttrs enableRocm {
   CUDACXX = "${cudaToolkit}/bin/nvcc";
   CUDAToolkit_ROOT = cudaToolkit;
 }) // {
-  inherit pname version src;
-  vendorHash = "sha256-Lj7CBvS51RqF63c01cOCgY7BCQeCKGu794qzb/S80C0=";
+  inherit pname version src vendorHash;
 
   nativeBuildInputs = [
     cmake
@@ -133,31 +145,20 @@ goBuild ((lib.optionalAttrs enableRocm {
     metalFrameworks;
 
   patches = [
-    # remove uses of `git` in the `go generate` script
-    # instead use `patch` where necessary
-    ./remove-git.patch
-    # replace a hardcoded use of `g++` with `$CXX`
-    ./replace-gcc.patch
-
-    # ollama's patches of llama.cpp's example server
-    # `ollama/llm/generate/gen_common.sh` -> "apply temporary patches until fix is upstream"
-    (preparePatch "01-cache.diff" "sha256-VDwu/iK6taBCyscpndQiOJ3eGqonnLVwmS2rJNMBVGU=")
-    (preparePatch "02-cudaleaks.diff" "sha256-nxsWgrePUMsZBWWQAjqVHWMJPzr1owH1zSJvUU7Q5pA=")
-    (preparePatch "03-load_exception.diff" "sha256-1DfNahFYYxqlx4E4pwMKQpL+XR0bibYnDFGt6dCL4TM=")
-    (preparePatch "04-locale.diff" "sha256-r5nHiP6yN/rQObRu2FZIPBKpKP9yByyZ6sSI2SKj6Do=")
-    (preparePatch "05-fix-clip-free.diff" "sha256-EFZ+QTtZCvstVxYgVdFKHsQqdkL98T0eXOEBOqCrlL4=")
-  ];
+    # disable uses of `git` in the `go generate` script
+    # ollama's build script assumes the source is a git repo, but nix removes the git directory
+    # this also disables necessary patches contained in `ollama/llm/patches/`
+    # those patches are added to `llamacppPatches`, and reapplied here in the patch phase
+    ./disable-git.patch
+  ] ++ llamacppPatches;
   postPatch = ''
-    # use a patch from the nix store in the `go generate` script
-    substituteInPlace llm/generate/gen_common.sh \
-      --subst-var-by cmakeIncludePatch '${./cmake-include.patch}'
-    # `ollama/llm/generate/gen_common.sh` -> "avoid duplicate main symbols when we link into the cgo binary"
-    substituteInPlace llm/llama.cpp/examples/server/server.cpp \
-      --replace-fail 'int main(' 'int __main('
+    # replace a hardcoded use of `g++` with `$CXX` so clang can be used on darwin
+    substituteInPlace llm/generate/gen_common.sh --replace-fail 'g++' '$CXX'
     # replace inaccurate version number with actual release version
     substituteInPlace version/version.go --replace-fail 0.0.0 '${version}'
   '';
   preBuild = ''
+    # disable uses of `git`, since nix removes the git directory
     export OLLAMA_SKIP_PATCHING=true
     # build llama.cpp libraries for ollama
     go generate ./...
@@ -168,9 +169,10 @@ goBuild ((lib.optionalAttrs enableRocm {
   '' + lib.optionalString (enableRocm || enableCuda) ''
     # expose runtime libraries necessary to use the gpu
     mv "$out/bin/ollama" "$out/bin/.ollama-unwrapped"
-    makeWrapper "$out/bin/.ollama-unwrapped" "$out/bin/ollama" \
-      --suffix LD_LIBRARY_PATH : '/run/opengl-driver/lib:${lib.makeLibraryPath runtimeLibs}' '' + lib.optionalString enableRocm ''\
-      --set-default HIP_PATH ${rocmPath}
+    makeWrapper "$out/bin/.ollama-unwrapped" "$out/bin/ollama" ${
+      lib.optionalString enableRocm
+        ''--set-default HIP_PATH '${rocmPath}' ''} \
+      --suffix LD_LIBRARY_PATH : '/run/opengl-driver/lib:${lib.makeLibraryPath runtimeLibs}'
   '';
 
   ldflags = [
@@ -191,9 +193,9 @@ goBuild ((lib.optionalAttrs enableRocm {
   };
 
   meta = {
-    changelog = "https://github.com/ollama/ollama/releases/tag/v${version}";
     description = "Get up and running with large language models locally";
-    homepage = "https://github.com/jmorganca/ollama";
+    homepage = "https://github.com/ollama/ollama";
+    changelog = "https://github.com/ollama/ollama/releases/tag/v${version}";
     license = licenses.mit;
     platforms = platforms.unix;
     mainProgram = "ollama";

pkgs/tools/misc/ollama/disable-git.patch

@@ -0,0 +1,20 @@
+--- a/llm/generate/gen_common.sh
++++ b/llm/generate/gen_common.sh
+@@ -65,6 +65,8 @@
+         echo 'add_subdirectory(../ext_server ext_server) # ollama' >>${LLAMACPP_DIR}/CMakeLists.txt
+     fi
+ 
++    return
++
+     if [ -n "$(ls -A ../patches/*.diff)" ]; then
+         # apply temporary patches until fix is upstream
+         for patch in ../patches/*.diff; do
+@@ -110,6 +112,8 @@
+ 
+ # Keep the local tree clean after we're done with the build
+ cleanup() {
++    return
++
+     (cd ${LLAMACPP_DIR}/ && git checkout CMakeLists.txt)
+ 
+     if [ -n "$(ls -A ../patches/*.diff)" ]; then

pkgs/tools/misc/ollama/remove-git.patch

@@ -1,21 +0,0 @@
---- a/llm/generate/gen_common.sh
-+++ b/llm/generate/gen_common.sh
-@@ -60,6 +60,9 @@
- }
- 
- apply_patches() {
-+    patch -i '@cmakeIncludePatch@' "${LLAMACPP_DIR}/examples/server/CMakeLists.txt"
-+    return
-+    
-     # Wire up our CMakefile
-     if ! grep ollama ${LLAMACPP_DIR}/examples/server/CMakeLists.txt; then
-         echo 'include (../../../ext_server/CMakeLists.txt) # ollama' >>${LLAMACPP_DIR}/examples/server/CMakeLists.txt
-@@ -113,6 +116,8 @@
- 
- # Keep the local tree clean after we're done with the build
- cleanup() {
-+    return
-+
-     (cd ${LLAMACPP_DIR}/examples/server/ && git checkout CMakeLists.txt server.cpp)
- 
-     if [ -n "$(ls -A ../patches/*.diff)" ]; then

pkgs/tools/misc/ollama/replace-gcc.patch

@@ -1,11 +0,0 @@
---- a/llm/generate/gen_common.sh
-+++ b/llm/generate/gen_common.sh
-@@ -86,7 +89,7 @@
-     cmake -S ${LLAMACPP_DIR} -B ${BUILD_DIR} ${CMAKE_DEFS}
-     cmake --build ${BUILD_DIR} ${CMAKE_TARGETS} -j8
-     mkdir -p ${BUILD_DIR}/lib/
--    g++ -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.${LIB_EXT} \
-+    $CXX -fPIC -g -shared -o ${BUILD_DIR}/lib/libext_server.${LIB_EXT} \
-         ${GCC_ARCH} \
-         ${WHOLE_ARCHIVE} ${BUILD_DIR}/examples/server/libext_server.a ${NO_WHOLE_ARCHIVE} \
-         ${BUILD_DIR}/common/libcommon.a \