Merge branch 'ggerganov:master' into master

rubra-ai · Mar 20, 2024 · 513fbf0 · 513fbf0
2 parents 177414d + d8b009a
commit 513fbf0
Show file tree

Hide file tree

Showing 69 changed files with 5,374 additions and 2,499 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -48,8 +48,8 @@ jobs:
           CC=gcc-8 make tests -j $(nproc)
           make test -j $(nproc)
 
-  ubuntu-latest-cmake:
-    runs-on: ubuntu-latest
+  ubuntu-focal-make-curl:
+    runs-on: ubuntu-20.04
 
     steps:
       - name: Clone
@@ -60,32 +60,19 @@ jobs:
         id: depends
         run: |
           sudo apt-get update
-          sudo apt-get install build-essential
+          sudo apt-get install build-essential gcc-8 libcurl4-openssl-dev
 
       - name: Build
-        id: cmake_build
-        run: |
-          mkdir build
-          cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON
-          cmake --build . --config Release -j $(nproc)
-
-      - name: Test
-        id: cmake_test
+        id: make_build
+        env:
+          LLAMA_FATAL_WARNINGS: 1
+          LLAMA_CURL: 1
         run: |
-          cd build
-          ctest -L main --verbose --timeout 900
+          CC=gcc-8 make -j $(nproc)
 
-  ubuntu-latest-cmake-sanitizer:
+  ubuntu-latest-cmake:
     runs-on: ubuntu-latest
 
-    continue-on-error: true
-
-    strategy:
-      matrix:
-        sanitizer: [ADDRESS, THREAD, UNDEFINED]
-        build_type: [Debug, Release]
-
     steps:
       - name: Clone
         id: checkout
@@ -102,15 +89,50 @@ jobs:
         run: |
           mkdir build
           cd build
-          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
-          cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
+          cmake .. -DLLAMA_FATAL_WARNINGS=ON
+          cmake --build . --config Release -j $(nproc)
 
       - name: Test
         id: cmake_test
         run: |
           cd build
           ctest -L main --verbose --timeout 900
 
+#  ubuntu-latest-cmake-sanitizer:
+#    runs-on: ubuntu-latest
+#
+#    continue-on-error: true
+#
+#    strategy:
+#      matrix:
+#        sanitizer: [ADDRESS, THREAD, UNDEFINED]
+#        build_type: [Debug, Release]
+#
+#    steps:
+#      - name: Clone
+#        id: checkout
+#        uses: actions/checkout@v3
+#
+#      - name: Dependencies
+#        id: depends
+#        run: |
+#          sudo apt-get update
+#          sudo apt-get install build-essential
+#
+#      - name: Build
+#        id: cmake_build
+#        run: |
+#          mkdir build
+#          cd build
+#          cmake .. -DLLAMA_FATAL_WARNINGS=ON -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON -DCMAKE_BUILD_TYPE=${{ matrix.build_type }}
+#          cmake --build . --config ${{ matrix.build_type }} -j $(nproc)
+#
+#      - name: Test
+#        id: cmake_test
+#        run: |
+#          cd build
+#          ctest -L main --verbose --timeout 900
+
   ubuntu-latest-cmake-mpi:
     runs-on: ubuntu-latest
 
@@ -333,6 +355,7 @@ jobs:
           mkdir build
           cd build
           cmake -G Xcode .. \
+            -DLLAMA_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
             -DLLAMA_BUILD_SERVER=OFF \
@@ -361,6 +384,7 @@ jobs:
           mkdir build
           cd build
           cmake -G Xcode .. \
+            -DLLAMA_METAL_EMBED_LIBRARY=ON \
             -DLLAMA_BUILD_EXAMPLES=OFF \
             -DLLAMA_BUILD_TESTS=OFF \
             -DLLAMA_BUILD_SERVER=OFF \

diff --git a/.github/workflows/close-issue.yml b/.github/workflows/close-issue.yml
@@ -0,0 +1,23 @@
+name: Close inactive issues
+on:
+  schedule:
+    - cron: "42 0 * * *"
+
+jobs:
+  close-issues:
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+    steps:
+      - uses: actions/stale@v5
+        with:
+          exempt-issue-labels: "refactor,help wanted,good first issue,research"
+          days-before-issue-stale: 30
+          days-before-issue-close: 14
+          stale-issue-label: "stale"
+          close-issue-message: "This issue was closed because it has been inactive for 14 days since being marked as stale."
+          days-before-pr-stale: -1
+          days-before-pr-close: -1
+          operations-per-run: 1000
+          repo-token: ${{ secrets.GITHUB_TOKEN }}
diff --git a/.github/workflows/server.yml b/.github/workflows/server.yml
@@ -24,18 +24,15 @@ jobs:
 
     strategy:
       matrix:
-        sanitizer: [ADDRESS, THREAD, UNDEFINED]
-        build_type: [Debug, Release]
+        # TODO: temporary disabled due to linux kernel issues
+        #sanitizer: [ADDRESS, THREAD, UNDEFINED]
+        sanitizer: [UNDEFINED]
+        build_type: [Debug]
         include:
           - build_type: Release
             sanitizer: ""
-        exclude:
-          - build_type: Release
-            sanitizer: ADDRESS
-          - build_type: Release
-            sanitizer: THREAD
-          - build_type: Release
-            sanitizer: UNDEFINED
+            disabled_on_pr: true
+      fail-fast: false # While -DLLAMA_SANITIZE_THREAD=ON is broken
 
     container:
       image: ubuntu:latest
@@ -60,7 +57,8 @@ jobs:
             cmake \
             python3-pip \
             wget \
-            language-pack-en
+            language-pack-en \
+            libcurl4-openssl-dev
 
       - name: Build
         id: cmake_build
@@ -70,6 +68,7 @@ jobs:
           cmake .. \
               -DLLAMA_NATIVE=OFF \
               -DLLAMA_BUILD_SERVER=ON \
+              -DLLAMA_CURL=ON \
               -DCMAKE_BUILD_TYPE=${{ matrix.build_type }} \
               -DLLAMA_SANITIZE_${{ matrix.sanitizer }}=ON ;
           cmake --build . --config ${{ matrix.build_type }} -j $(nproc) --target server
@@ -81,13 +80,14 @@ jobs:
 
       - name: Tests
         id: server_integration_tests
+        if: ${{ !matrix.disabled_on_pr || !github.event.pull_request }}
         run: |
           cd examples/server/tests
           PORT=8888 ./tests.sh
 
       - name: Slow tests
         id: server_integration_tests_slow
-        if: ${{ github.event.schedule != '' && matrix.build_type == 'Release' || github.event.inputs.slow_tests == 'true' }}
+        if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
         run: |
           cd examples/server/tests
           PORT=8888 ./tests.sh --stop --no-skipped --no-capture --tags slow
@@ -103,12 +103,21 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: libCURL
+        id: get_libcurl
+        env:
+          CURL_VERSION: 8.6.0_6
+        run: |
+          curl.exe -o $env:RUNNER_TEMP/curl.zip -L "https://curl.se/windows/dl-${env:CURL_VERSION}/curl-${env:CURL_VERSION}-win64-mingw.zip"
+          mkdir $env:RUNNER_TEMP/libcurl
+          tar.exe -xvf $env:RUNNER_TEMP/curl.zip --strip-components=1 -C $env:RUNNER_TEMP/libcurl
+
       - name: Build
         id: cmake_build
         run: |
           mkdir build
           cd build
-          cmake ..  -DLLAMA_BUILD_SERVER=ON -DCMAKE_BUILD_TYPE=Release ;
+          cmake .. -DLLAMA_CURL=ON -DCURL_LIBRARY="$env:RUNNER_TEMP/libcurl/lib/libcurl.dll.a" -DCURL_INCLUDE_DIR="$env:RUNNER_TEMP/libcurl/include"
           cmake --build . --config Release -j ${env:NUMBER_OF_PROCESSORS} --target server
 
       - name: Python setup
@@ -122,15 +131,21 @@ jobs:
         run: |
           pip install -r examples/server/tests/requirements.txt
 
+      - name: Copy Libcurl
+        id: prepare_libcurl
+        run: |
+          cp $env:RUNNER_TEMP/libcurl/bin/libcurl-x64.dll ./build/bin/Release/libcurl-x64.dll
+
       - name: Tests
         id: server_integration_tests
+        if: ${{ !matrix.disabled_on_pr || !github.event.pull_request }}
         run: |
           cd examples/server/tests
           behave.exe --summary --stop --no-capture --exclude 'issues|wrong_usages|passkey' --tags llama.cpp
 
       - name: Slow tests
         id: server_integration_tests_slow
-        if: ${{ github.event.schedule != '' || github.event.inputs.slow_tests == 'true' }}
+        if: ${{ (github.event.schedule || github.event.inputs.slow_tests == 'true') && matrix.build_type == 'Release' }}
         run: |
           cd examples/server/tests
           behave.exe --stop --no-skipped --no-capture --tags slow
diff --git a/.gitignore b/.gitignore
@@ -25,6 +25,8 @@
 .vscode/
 .idea/
 
+ggml-metal-embed.metal
+
 lcov-report/
 gcovr-report/
 

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -99,6 +99,7 @@ option(LLAMA_CUDA_F16                        "llama: use 16 bit floats for some
 set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for Q2_K/Q6_K")
 set(LLAMA_CUDA_PEER_MAX_BATCH_SIZE "128" CACHE STRING
                                              "llama: max. batch size for using peer access")
+option(LLAMA_CURL                            "llama: use libcurl to download model from an URL" OFF)
 option(LLAMA_HIPBLAS                         "llama: use hipBLAS"                               OFF)
 option(LLAMA_HIP_UMA                         "llama: use HIP unified memory architecture"       OFF)
 option(LLAMA_CLBLAST                         "llama: use CLBlast"                               OFF)
@@ -118,6 +119,7 @@ option(LLAMA_SYCL                            "llama: use SYCL"
 option(LLAMA_SYCL_F16                        "llama: use 16 bit floats for sycl calculations"   OFF)
 set(LLAMA_SYCL_TARGET   "INTEL" CACHE STRING "llama: sycl target device")
 option(LLAMA_CPU_HBM                         "llama: use memkind for CPU HBM"                   OFF)
+set(LLAMA_SCHED_MAX_COPIES  "4" CACHE STRING "llama: max input copies for pipeline parallelism")
 
 option(LLAMA_BUILD_TESTS                     "llama: build tests"    ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_EXAMPLES                  "llama: build examples" ${LLAMA_STANDALONE})
@@ -147,6 +149,8 @@ set(THREADS_PREFER_PTHREAD_FLAG ON)
 find_package(Threads REQUIRED)
 include(CheckCXXCompilerFlag)
 
+add_compile_definitions(GGML_SCHED_MAX_COPIES=${LLAMA_SCHED_MAX_COPIES})
+
 # enable libstdc++ assertions for debug builds
 if (CMAKE_SYSTEM_NAME MATCHES "Linux")
     add_compile_definitions($<$<CONFIG:Debug>:_GLIBCXX_ASSERTIONS>)
@@ -197,9 +201,6 @@ if (LLAMA_METAL)
         add_compile_definitions(GGML_METAL_NDEBUG)
     endif()
 
-    # get full path to the file
-    #add_compile_definitions(GGML_METAL_DIR_KERNELS="${CMAKE_CURRENT_SOURCE_DIR}/")
-
     # copy ggml-common.h and ggml-metal.metal to bin directory
     configure_file(ggml-common.h    ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h    COPYONLY)
     configure_file(ggml-metal.metal ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal COPYONLY)
@@ -208,53 +209,62 @@ if (LLAMA_METAL)
         enable_language(ASM)
         add_compile_definitions(GGML_METAL_EMBED_LIBRARY)
 
+        set(METALLIB_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/ggml-common.h")
         set(METALLIB_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
+
         file(MAKE_DIRECTORY "${CMAKE_BINARY_DIR}/autogenerated")
-        set(EMBED_METALLIB_ASSEMBLY "${CMAKE_BINARY_DIR}/autogenerated/ggml-embed-metallib.s")
+
+        # merge ggml-common.h and ggml-metal.metal into a single file
+        set(METALLIB_EMBED_ASM    "${CMAKE_BINARY_DIR}/autogenerated/ggml-metal-embed.s")
+        set(METALLIB_SOURCE_EMBED "${CMAKE_BINARY_DIR}/autogenerated/ggml-metal-embed.metal")
 
         add_custom_command(
-            OUTPUT ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo ".section __DATA,__ggml_metallib" > ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo ".globl _ggml_metallib_start" >> ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo "_ggml_metallib_start:" >> ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo ".incbin \\\"${METALLIB_SOURCE}\\\"" >> ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo ".globl _ggml_metallib_end" >> ${EMBED_METALLIB_ASSEMBLY}
-            COMMAND echo "_ggml_metallib_end:" >> ${EMBED_METALLIB_ASSEMBLY}
-            DEPENDS ${METALLIB_SOURCE}
+            OUTPUT ${METALLIB_EMBED_ASM}
+            COMMAND echo "Embedding Metal library"
+            COMMAND sed -e '/\#include \"ggml-common.h\"/r ${METALLIB_COMMON}' -e '/\#include \"ggml-common.h\"/d' < ${METALLIB_SOURCE} > ${METALLIB_SOURCE_EMBED}
+            COMMAND echo ".section __DATA,__ggml_metallib"          >  ${METALLIB_EMBED_ASM}
+            COMMAND echo ".globl _ggml_metallib_start"              >> ${METALLIB_EMBED_ASM}
+            COMMAND echo "_ggml_metallib_start:"                    >> ${METALLIB_EMBED_ASM}
+            COMMAND echo ".incbin \\\"${METALLIB_SOURCE_EMBED}\\\"" >> ${METALLIB_EMBED_ASM}
+            COMMAND echo ".globl _ggml_metallib_end"                >> ${METALLIB_EMBED_ASM}
+            COMMAND echo "_ggml_metallib_end:"                      >> ${METALLIB_EMBED_ASM}
+            DEPENDS ggml-metal.metal ggml-common.h
             COMMENT "Generate assembly for embedded Metal library"
         )
 
-        set(GGML_SOURCES_METAL ${GGML_SOURCES_METAL} ${EMBED_METALLIB_ASSEMBLY})
-    endif()
-
-    if (LLAMA_METAL_SHADER_DEBUG)
-        # custom command to do the following:
-        #   xcrun -sdk macosx metal    -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
-        #   xcrun -sdk macosx metallib                   ggml-metal.air   -o default.metallib
-        #
-        # note: this is the only way I found to disable fast-math in Metal. it's ugly, but at least it works
-        #       disabling fast math is needed in order to pass tests/test-backend-ops
-        # note: adding -fno-inline fixes the tests when using MTL_SHADER_VALIDATION=1
-        # note: unfortunately, we have to call it default.metallib instead of ggml.metallib
-        #       ref: https://github.com/ggerganov/whisper.cpp/issues/1720
-        set(XC_FLAGS -fno-fast-math -fno-inline -g)
-        if (LLAMA_QKK_64)
-            set(XC_FLAGS ${XC_FLAGS} -DQK_K=64)
+        set(GGML_SOURCES_METAL ${GGML_SOURCES_METAL} ${METALLIB_EMBED_ASM})
+    else()
+        if (LLAMA_METAL_SHADER_DEBUG)
+            # custom command to do the following:
+            #   xcrun -sdk macosx metal    -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
+            #   xcrun -sdk macosx metallib                   ggml-metal.air   -o default.metallib
+            #
+            # note: this is the only way I found to disable fast-math in Metal. it's ugly, but at least it works
+            #       disabling fast math is needed in order to pass tests/test-backend-ops
+            # note: adding -fno-inline fixes the tests when using MTL_SHADER_VALIDATION=1
+            # note: unfortunately, we have to call it default.metallib instead of ggml.metallib
+            #       ref: https://github.com/ggerganov/whisper.cpp/issues/1720
+            set(XC_FLAGS -fno-fast-math -fno-inline -g)
+        else()
+            set(XC_FLAGS -O3)
         endif()
 
         add_custom_command(
             OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
             COMMAND xcrun -sdk macosx metal    ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
             COMMAND xcrun -sdk macosx metallib                ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air   -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
-            DEPENDS ggml-metal.metal
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
+            COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal
+            DEPENDS ggml-metal.metal ggml-common.h
             COMMENT "Compiling Metal kernels"
-        )
+            )
 
         add_custom_target(
             ggml-metal ALL
             DEPENDS ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
-        )
-    endif()
+            )
+    endif() # LLAMA_METAL_EMBED_LIBRARY
 
     set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS}
         ${FOUNDATION_LIBRARY}