google / sentencepiece

Unsupervised text tokenizer for Neural Network-based text generation.
Apache License 2.0
10.07k stars 1.16k forks source link

Runtime error on iOS #1010

Closed l3utterfly closed 1 month ago

l3utterfly commented 4 months ago

After building for iOS 17, I get this error when loading a model:

sentencepiece::SentencePieceProcessor srcSPM, destSPM;
    auto loadSrcStatus = srcSPM.Load(modelPathStr + "/source.spm");
    auto loadDestStatus = destSPM.Load(modelPathStr + "/target.spm");

Error:

[libprotobuf FATAL /Users/runner/work/1/b/ios_framework/intermediates/iphoneos_arm64/Release/_deps/protobuf-src/src/google/protobuf/stubs/common.cc:83] This program was compiled against version 3.14.0 of the Protocol Buffer runtime library, which is not compatible with the installed version (3.21.12).  Contact the program author for an update.  If you compiled the program yourself, make sure that your headers are from the same version of Protocol Buffers as your link-time library.  (Version verification failed in "/Users/username/Documents/myprogram/cpp/sentencepiece/src/builtin_pb/sentencepiece_model.pb.cc".)
taku910 commented 4 months ago

The error message describes the root cause. incompatible protobuf library are used in .cc and .h. I guess this library is built with non-standard way. Could you provide the details of how this SentencePiece library was built?

l3utterfly commented 4 months ago

I cloned the master branch, and generated the xcodeproj file via: cmake .. -G Xcode -DCMAKE_TOOLCHAIN_FILE=../cmake/ios.toolchain.cmake -DPLATFORM=OS64

That's it. I'm linking against the static library in my iOS project.

l3utterfly commented 4 months ago

I looked through the cmake file, by default sentencepiece uses the pre-built generated protobuf. It seems SPM_PROTOBUF_PROVIDER=internal by default. Is the generated protobuf files commited to this repo generated by protobuf 3.14?

taku910 commented 3 months ago

I looked through the cmake file, by default sentencepiece uses the pre-built generated protobuf. It seems SPM_PROTOBUF_PROVIDER=internal by default. Is the generated protobuf files commited to this repo generated by protobuf 3.14?

Right. SPM_PROTOBUF_PROVIDER=package will allow you to build the protobuf already installed on your system.

l3utterfly commented 3 months ago

@taku910 I'm not sure if I understand correctly. If I use SPM_PROTOBUF_PROVIDER=internal, shouldn't the built sentencepiece library use internally provided protobuf library included in this project? Why would it look for protobuf 3.14 that's installed in the iPhone?

taku910 commented 3 months ago

It is possible that the linkage is mismatched: in the case of internal, sentencepiece is linking to the internal library. On the other hand, if the parent project (iPhone) is trying to link another version of the sentencepiece, duplication will occur. SPM_PROTOBUF_PROVIDER=package does only find the sentencepiece package that is installed as a library on the standard linux system., i.e., so, h, *a files are all available. Probably the libraries required to build sentence pieces are not available.

iprovalo commented 3 months ago

@l3utterfly were you able to find a work-around? I am seeing the same issue.

Thank you!

l3utterfly commented 3 months ago

@iprovalo I believe the issue was with iOS minimum version. I recently upgraded to support minimum version iOS 15.5 and the problem went away

iprovalo commented 3 months ago

@l3utterfly thank you very much!

iprovalo commented 3 months ago

@l3utterfly @taku910

I am still running into this issue if I use a static xcframework.

The way I am building it is by first adding this modification to CMakeList.txt (to avoid Unknown CMake command "set_xcode_property"):

if(CMAKE_SYSTEM_NAME STREQUAL "iOS")
    macro(set_xcode_property TARGET XCODE_PROPERTY XCODE_VALUE)
        set_property(TARGET ${TARGET} PROPERTY
                XCODE_ATTRIBUTE_${XCODE_PROPERTY} ${XCODE_VALUE})
    endmacro(set_xcode_property)
endif()

Then creating both a dynamic and static xcframework like so (I tried the deployment target 15.5, same results):

set -e

TEAM_ID="XXX"
DEPLOYMENT_CONFIG=Release
ARCH="arm64;x86_64"
DEPLOYMENT_TARGET="15.5"

SOURCE_DIR="src/"
SOURCE_3RD_PARTY_DIR="third_party/"
BUILD_DIR=build-ios
DEST_HEADER_DIR="${BUILD_DIR}/include"

rm -Rf "${BUILD_DIR}"

cmake -B${BUILD_DIR} -GXcode -DCMAKE_SYSTEM_NAME=iOS \
    -DCMAKE_OSX_DEPLOYMENT_TARGET=$DEPLOYMENT_TARGET \
    -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM="$TEAM_ID" \
    -DCMAKE_OSX_ARCHITECTURES=$ARCH

mkdir -p "${DEST_HEADER_DIR}"

cd ${BUILD_DIR}

xcodebuild -configuration $DEPLOYMENT_CONFIG -sdk iphoneos -target sentencepiece
xcodebuild -configuration $DEPLOYMENT_CONFIG -sdk iphoneos -target sentencepiece-static

xcodebuild -configuration $DEPLOYMENT_CONFIG -sdk iphonesimulator -target sentencepiece
xcodebuild -configuration $DEPLOYMENT_CONFIG -sdk iphonesimulator -target sentencepiece-static

cd ../

find "${SOURCE_DIR}" -name '*.h' | while read -r header; do
    relative_dir=$(dirname "${header#${SOURCE_DIR}/}")
    echo "copying $header to ${DEST_HEADER_DIR}/$relative_dir"
    mkdir -p "${DEST_HEADER_DIR}/${relative_dir}"
    cp "${header}" "${DEST_HEADER_DIR}/${relative_dir}"
done
echo "Headers copied to ${DEST_HEADER_DIR}"

find "${SOURCE_3RD_PARTY_DIR}" -name '*.h' | while read -r header; do
    relative_dir=$(dirname "${header#${SOURCE_3RD_PARTY_DIR}/}")
    echo "copying $header to ${DEST_HEADER_DIR}/${SOURCE_3RD_PARTY_DIR}/$relative_dir"
    mkdir -p "${DEST_HEADER_DIR}/${SOURCE_3RD_PARTY_DIR}/${relative_dir}"
    cp "${header}" "${DEST_HEADER_DIR}/${SOURCE_3RD_PARTY_DIR}/${relative_dir}"
done
echo "Third party Headers copied to ${DEST_HEADER_DIR}"

DEVICE_DIR=${BUILD_DIR}/src/${DEPLOYMENT_CONFIG}-iphoneos
SIMULATOR_DIR=${BUILD_DIR}/src/${DEPLOYMENT_CONFIG}-iphonesimulator
XCFRAMEWORK_DIR=${BUILD_DIR}/${DEPLOYMENT_CONFIG}-universal

xcodebuild -create-xcframework \
    -library ${DEVICE_DIR}/libsentencepiece.a \
    -headers ${DEST_HEADER_DIR} \
    -library ${SIMULATOR_DIR}/libsentencepiece.a \
    -headers ${DEST_HEADER_DIR} \
    -output ${XCFRAMEWORK_DIR}-static/sentencepiece.xcframework

xcodebuild -create-xcframework \
    -library ${DEVICE_DIR}/libsentencepiece.0.0.0.dylib \
    -headers ${DEST_HEADER_DIR} \
    -library ${SIMULATOR_DIR}/libsentencepiece.0.0.0.dylib \
    -headers ${DEST_HEADER_DIR} \
    -output ${XCFRAMEWORK_DIR}/sentencepiece.xcframework
iprovalo commented 3 months ago

I ended up wrapping the static sentencepiece lib in an XCode Framework, and it worked.