From d648590ae51f0ba0081adc5146088d53b70a3371 Mon Sep 17 00:00:00 2001 From: vansangpfiev Date: Fri, 2 Aug 2024 10:05:36 +0700 Subject: [PATCH 1/3] fix: race condition issue for server example --- examples/server/server.cc | 34 +++++++++++++++++++++++----------- 1 file changed, 23 insertions(+), 11 deletions(-) diff --git a/examples/server/server.cc b/examples/server/server.cc index 274fe69..acae5b7 100644 --- a/examples/server/server.cc +++ b/examples/server/server.cc @@ -49,6 +49,18 @@ class Server { }; }; +struct SyncJsonReader { + public: + void Parse(const std::string& document, Json::Value& root) { + std::lock_guard l(m); + r.parse(document, root); + } + + private: + Json::Reader r; + std::mutex m; +}; + std::function shutdown_handler; std::atomic_flag is_terminating = ATOMIC_FLAG_INIT; @@ -78,7 +90,7 @@ int main(int argc, char** argv) { } Server server; - Json::Reader r; + SyncJsonReader r; auto svr = std::make_unique(); if (!svr->bind_to_port(hostname, port)) { @@ -127,7 +139,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); server.engine_->LoadModel( req_body, [&server, &resp](Json::Value status, Json::Value res) { resp.set_content(res.toStyledString().c_str(), @@ -141,7 +153,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); server.engine_->UnloadModel( req_body, [&server, &resp](Json::Value status, Json::Value res) { resp.set_content(res.toStyledString().c_str(), @@ -155,7 +167,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); bool is_stream = (*req_body).get("stream", false).asBool(); // This is an async call, need to use queue auto q = std::make_shared(); @@ -175,7 +187,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); // This is an async call, need to use queue SyncQueue q; server.engine_->HandleEmbedding( @@ -190,7 +202,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); server.engine_->GetModelStatus( req_body, [&server, &resp](Json::Value status, Json::Value res) { resp.set_content(res.toStyledString().c_str(), @@ -204,7 +216,7 @@ int main(int argc, char** argv) { resp.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin")); auto req_body = std::make_shared(); - r.parse(req.body, *req_body); + r.Parse(req.body, *req_body); server.engine_->GetModels( req_body, [&server, &resp](Json::Value status, Json::Value res) { resp.set_content(res.toStyledString().c_str(), @@ -222,10 +234,10 @@ int main(int argc, char** argv) { svr->Get("/models", handle_get_running_models); std::atomic running = true; svr->Delete("/destroy", - [&](const httplib::Request& req, httplib::Response& resp) { - LOG_INFO << "Received Stop command"; - running = false; - }); + [&](const httplib::Request& req, httplib::Response& resp) { + LOG_INFO << "Received Stop command"; + running = false; + }); LOG_INFO << "HTTP server listening: " << hostname << ":" << port; svr->new_task_queue = [] { From 1517d1538f20f0ea7d2a38a679544f00b7d33c5b Mon Sep 17 00:00:00 2001 From: nguyenhoangthuan99 Date: Fri, 2 Aug 2024 05:17:23 +0000 Subject: [PATCH 2/3] test cuda build noavx --- .../workflows/template-quality-gate-pr.yml | 334 +++++++++--------- 1 file changed, 171 insertions(+), 163 deletions(-) diff --git a/.github/workflows/template-quality-gate-pr.yml b/.github/workflows/template-quality-gate-pr.yml index 7cd406c..1afd67f 100644 --- a/.github/workflows/template-quality-gate-pr.yml +++ b/.github/workflows/template-quality-gate-pr.yml @@ -28,66 +28,74 @@ jobs: fail-fast: false matrix: include: + # - os: "linux" + # name: "amd64-avx2" + # runs-on: "ubuntu-20-04" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: true + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-noavx" + # runs-on: "ubuntu-20-04" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-avx" + # runs-on: "ubuntu-20-04" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-avx512" + # runs-on: "ubuntu-20-04" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-vulkan" + # runs-on: "ubuntu-22-04" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_VULKAN=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: true + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-avx2-cuda-11-7" + # runs-on: "ubuntu-20-04-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-avx-cuda-11-7" + # runs-on: "ubuntu-20-04-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" + # - os: "linux" + # name: "amd64-avx512-cuda-11-7" + # runs-on: "ubuntu-20-04-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: "/tmp/sccache.conf" - os: "linux" - name: "amd64-avx2" - runs-on: "ubuntu-20-04" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: true - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-noavx" - runs-on: "ubuntu-20-04" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-avx" - runs-on: "ubuntu-20-04" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-avx512" - runs-on: "ubuntu-20-04" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-vulkan" - runs-on: "ubuntu-22-04" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_VULKAN=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: true - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-avx2-cuda-11-7" - runs-on: "ubuntu-20-04-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-avx-cuda-11-7" - runs-on: "ubuntu-20-04-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: "/tmp/sccache.conf" - - os: "linux" - name: "amd64-avx512-cuda-11-7" - runs-on: "ubuntu-20-04-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + name: "amd64-noavx-cuda-12-0" + runs-on: "ubuntu-20-04-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_FMA=OFF -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" run-e2e: false vulkan: false sccache: true @@ -116,110 +124,110 @@ jobs: vulkan: false sccache: true sccache-conf-path: "/tmp/sccache.conf" - - os: "mac" - name: "amd64" - runs-on: "macos-13" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL=OFF" - run-e2e: true - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "mac" - name: "arm64" - runs-on: "macos-silicon" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL_EMBED_LIBRARY=ON" - run-e2e: true - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-avx2" - runs-on: "windows-2019" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - run-e2e: true - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-noavx" - runs-on: "windows-2019" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - run-e2e: false - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-avx" - runs-on: "windows-2019" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - run-e2e: true - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-avx512" - runs-on: "windows-2019" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - run-e2e: false - vulkan: false - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-vulkan" - runs-on: "windows-2019" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - run-e2e: false - vulkan: true - sccache: false - sccache-conf-path: "" - - os: "windows" - name: "amd64-avx2-cuda-12-0" - runs-on: "windows-cuda-12-0" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' - - os: "windows" - name: "amd64-avx-cuda-12-0" - runs-on: "windows-cuda-12-0" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' - - os: "windows" - name: "amd64-avx512-cuda-12-0" - runs-on: "windows-cuda-12-0" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' - - os: "windows" - name: "amd64-avx2-cuda-11-7" - runs-on: "windows-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' - - os: "windows" - name: "amd64-avx-cuda-11-7" - runs-on: "windows-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' - - os: "windows" - name: "amd64-avx512-cuda-11-7" - runs-on: "windows-cuda-11-7" - cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - run-e2e: false - vulkan: false - sccache: true - sccache-conf-path: 'C:\sccache.conf' + # - os: "mac" + # name: "amd64" + # runs-on: "macos-13" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL=OFF" + # run-e2e: true + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "mac" + # name: "arm64" + # runs-on: "macos-silicon" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL_EMBED_LIBRARY=ON" + # run-e2e: true + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-avx2" + # runs-on: "windows-2019" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + # run-e2e: true + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-noavx" + # runs-on: "windows-2019" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + # run-e2e: false + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-avx" + # runs-on: "windows-2019" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + # run-e2e: true + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-avx512" + # runs-on: "windows-2019" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + # run-e2e: false + # vulkan: false + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-vulkan" + # runs-on: "windows-2019" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + # run-e2e: false + # vulkan: true + # sccache: false + # sccache-conf-path: "" + # - os: "windows" + # name: "amd64-avx2-cuda-12-0" + # runs-on: "windows-cuda-12-0" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' + # - os: "windows" + # name: "amd64-avx-cuda-12-0" + # runs-on: "windows-cuda-12-0" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' + # - os: "windows" + # name: "amd64-avx512-cuda-12-0" + # runs-on: "windows-cuda-12-0" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' + # - os: "windows" + # name: "amd64-avx2-cuda-11-7" + # runs-on: "windows-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' + # - os: "windows" + # name: "amd64-avx-cuda-11-7" + # runs-on: "windows-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' + # - os: "windows" + # name: "amd64-avx512-cuda-11-7" + # runs-on: "windows-cuda-11-7" + # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + # run-e2e: false + # vulkan: false + # sccache: true + # sccache-conf-path: 'C:\sccache.conf' steps: - name: Clone From 61416e6cbb186ff1e323024381c53a7cd925e620 Mon Sep 17 00:00:00 2001 From: nguyenhoangthuan99 Date: Fri, 2 Aug 2024 11:15:24 +0000 Subject: [PATCH 3/3] add build cuda without avx --- .github/workflows/build.yml | 32 ++ .github/workflows/nightly-build.yml | 32 ++ .../workflows/template-quality-gate-pr.yml | 360 ++++++++++-------- .../template-quality-gate-submodule.yml | 32 ++ 4 files changed, 288 insertions(+), 168 deletions(-) diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml index 6ccb8d5..677ac2e 100644 --- a/.github/workflows/build.yml +++ b/.github/workflows/build.yml @@ -101,6 +101,14 @@ jobs: vulkan: true sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-11-7" runs-on: "ubuntu-20-04-cuda-11-7" @@ -125,6 +133,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-12-0" + runs-on: "ubuntu-20-04-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-12-0" runs-on: "ubuntu-20-04-cuda-12-0" @@ -205,6 +221,14 @@ jobs: vulkan: true sccache: false sccache-conf-path: "" + - os: "windows" + name: "amd64-noavx-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-12-0" runs-on: "windows-cuda-12-0" @@ -229,6 +253,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-noavx-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-11-7" runs-on: "windows-cuda-11-7" diff --git a/.github/workflows/nightly-build.yml b/.github/workflows/nightly-build.yml index da8f0ab..b47f5e0 100644 --- a/.github/workflows/nightly-build.yml +++ b/.github/workflows/nightly-build.yml @@ -101,6 +101,14 @@ jobs: vulkan: true sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-11-7" runs-on: "ubuntu-20-04-cuda-11-7" @@ -125,6 +133,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-12-0" + runs-on: "ubuntu-20-04-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-12-0" runs-on: "ubuntu-20-04-cuda-12-0" @@ -205,6 +221,14 @@ jobs: vulkan: true sccache: false sccache-conf-path: "" + - os: "windows" + name: "amd64-noavx-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-12-0" runs-on: "windows-cuda-12-0" @@ -229,6 +253,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-noavx-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{needs.create-draft-release.outputs.version}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-11-7" runs-on: "windows-cuda-11-7" diff --git a/.github/workflows/template-quality-gate-pr.yml b/.github/workflows/template-quality-gate-pr.yml index 1afd67f..51cc03e 100644 --- a/.github/workflows/template-quality-gate-pr.yml +++ b/.github/workflows/template-quality-gate-pr.yml @@ -28,70 +28,78 @@ jobs: fail-fast: false matrix: include: - # - os: "linux" - # name: "amd64-avx2" - # runs-on: "ubuntu-20-04" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: true - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-noavx" - # runs-on: "ubuntu-20-04" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-avx" - # runs-on: "ubuntu-20-04" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-avx512" - # runs-on: "ubuntu-20-04" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-vulkan" - # runs-on: "ubuntu-22-04" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_VULKAN=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: true - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-avx2-cuda-11-7" - # runs-on: "ubuntu-20-04-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-avx-cuda-11-7" - # runs-on: "ubuntu-20-04-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" - # - os: "linux" - # name: "amd64-avx512-cuda-11-7" - # runs-on: "ubuntu-20-04-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx2" + runs-on: "ubuntu-20-04" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: true + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx" + runs-on: "ubuntu-20-04" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx" + runs-on: "ubuntu-20-04" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx512" + runs-on: "ubuntu-20-04" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-vulkan" + runs-on: "ubuntu-22-04" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_VULKAN=ON -DGGML_NATIVE=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: true + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx2-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-avx512-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-noavx-cuda-12-0" runs-on: "ubuntu-20-04-cuda-12-0" @@ -124,110 +132,126 @@ jobs: vulkan: false sccache: true sccache-conf-path: "/tmp/sccache.conf" - # - os: "mac" - # name: "amd64" - # runs-on: "macos-13" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL=OFF" - # run-e2e: true - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "mac" - # name: "arm64" - # runs-on: "macos-silicon" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL_EMBED_LIBRARY=ON" - # run-e2e: true - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-avx2" - # runs-on: "windows-2019" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - # run-e2e: true - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-noavx" - # runs-on: "windows-2019" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - # run-e2e: false - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-avx" - # runs-on: "windows-2019" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - # run-e2e: true - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-avx512" - # runs-on: "windows-2019" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - # run-e2e: false - # vulkan: false - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-vulkan" - # runs-on: "windows-2019" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" - # run-e2e: false - # vulkan: true - # sccache: false - # sccache-conf-path: "" - # - os: "windows" - # name: "amd64-avx2-cuda-12-0" - # runs-on: "windows-cuda-12-0" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' - # - os: "windows" - # name: "amd64-avx-cuda-12-0" - # runs-on: "windows-cuda-12-0" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' - # - os: "windows" - # name: "amd64-avx512-cuda-12-0" - # runs-on: "windows-cuda-12-0" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' - # - os: "windows" - # name: "amd64-avx2-cuda-11-7" - # runs-on: "windows-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' - # - os: "windows" - # name: "amd64-avx-cuda-11-7" - # runs-on: "windows-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' - # - os: "windows" - # name: "amd64-avx512-cuda-11-7" - # runs-on: "windows-cuda-11-7" - # cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" - # run-e2e: false - # vulkan: false - # sccache: true - # sccache-conf-path: 'C:\sccache.conf' + - os: "mac" + name: "amd64" + runs-on: "macos-13" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL=OFF" + run-e2e: true + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "mac" + name: "arm64" + runs-on: "macos-silicon" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_METAL_EMBED_LIBRARY=ON" + run-e2e: true + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-avx2" + runs-on: "windows-2019" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + run-e2e: true + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-noavx" + runs-on: "windows-2019" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DBUILD_SHARED_LIBS=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + run-e2e: false + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-avx" + runs-on: "windows-2019" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + run-e2e: true + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-avx512" + runs-on: "windows-2019" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_AVX512=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + run-e2e: false + vulkan: false + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-vulkan" + runs-on: "windows-2019" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_VULKAN=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER=cl -DCMAKE_C_COMPILER=cl -GNinja" + run-e2e: false + vulkan: true + sccache: false + sccache-conf-path: "" + - os: "windows" + name: "amd64-noavx-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx2-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx512-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-noavx-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx2-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-avx512-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX512=ON -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' steps: - name: Clone diff --git a/.github/workflows/template-quality-gate-submodule.yml b/.github/workflows/template-quality-gate-submodule.yml index 1ea3a72..03859ed 100644 --- a/.github/workflows/template-quality-gate-submodule.yml +++ b/.github/workflows/template-quality-gate-submodule.yml @@ -68,6 +68,14 @@ jobs: vulkan: true sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-11-7" + runs-on: "ubuntu-20-04-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_FMA=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_AVX2=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-11-7" runs-on: "ubuntu-20-04-cuda-11-7" @@ -92,6 +100,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: "/tmp/sccache.conf" + - os: "linux" + name: "amd64-noavx-cuda-12-0" + runs-on: "ubuntu-20-04-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_FMA=OFF -DGGML_AVX2=OFF -DBUILD_SHARED_LIBS=OFF -DGGML_NATIVE=OFF -DGGML_CUDA=ON -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_BUILD_TYPE='Release' -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: "/tmp/sccache.conf" - os: "linux" name: "amd64-avx2-cuda-12-0" runs-on: "ubuntu-20-04-cuda-12-0" @@ -172,6 +188,14 @@ jobs: vulkan: true sccache: false sccache-conf-path: "" + - os: "windows" + name: "amd64-noavx-cuda-12-0" + runs-on: "windows-cuda-12-0" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-12-0" runs-on: "windows-cuda-12-0" @@ -196,6 +220,14 @@ jobs: vulkan: false sccache: true sccache-conf-path: 'C:\sccache.conf' + - os: "windows" + name: "amd64-noavx-cuda-11-7" + runs-on: "windows-cuda-11-7" + cmake-flags: "-DCORTEXLLAMA_VERSION=${{github.event.pull_request.head.sha}} -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DGGML_NATIVE=OFF -DLLAMA_BUILD_SERVER=ON -DGGML_CUDA=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_BUILD_TYPE='Release' -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache -DCMAKE_CUDA_COMPILER_LAUNCHER=sccache -GNinja" + run-e2e: false + vulkan: false + sccache: true + sccache-conf-path: 'C:\sccache.conf' - os: "windows" name: "amd64-avx2-cuda-11-7" runs-on: "windows-cuda-11-7"