-
Notifications
You must be signed in to change notification settings - Fork 65
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changed Makefile to support TF2.6.1. #66
base: master
Are you sure you want to change the base?
Changed Makefile to support TF2.6.1. #66
Conversation
NobuoTsukamoto
commented
Apr 7, 2024
- Makefile compatible with TF2.6.1.
- Fixed undefined symbol issue.
- Missing symbol after compilation #28
- libedgetpu not working on raspberrypi4 NobuoTsukamoto/meta-tensorflow-lite#63
@NobuoTsukamoto could you update this for TFv2.16.1 ? |
@stakach Are any build errors occurring? |
This should be indeed pulled, as the Makefile in its current form is outdated. That said, when I try to build it, I have the following error, despite having
|
Please attach all build logs.
Also check the flatc version and installation path.
|
flatc --version
=> flatc version 2.0.8 on the latest version of ubuntu
|
Yes, flatc is in version 2.0.8 for me as well. Using pip to install the proper python package doesn't change the actual version of flatc. |
This worked to install it git clone https://github.com/google/flatbuffers.git
cd flatbuffers
git checkout tags/v23.5.26
cmake -G "Unix Makefiles" \
-DCMAKE_BUILD_TYPE=Release \
-DCMAKE_INSTALL_PREFIX=/usr/local \
-DFLATBUFFERS_BUILD_SHAREDLIB:BOOL=ON
make
sudo make install although the versioning is just based on release date and not semver so probably compatible with the 24 versions |
TF requires a specific version of flatbuffers depending on the TF release as well. For TF 2.16.1 it's 23.5.26. From: https://github.com/tensorflow/tensorflow/blob/v2.16.1/tensorflow/lite/schema/schema_generated.h
|
In any case, the README.md file should reflect that if someone wants to compile with Makefile, you need a specific version of flatbuffers as indicated (including compilation). |
the other issue is that manually building flatbuffers only installs a static lib but the libedgetpu build expects a shared lib - is there any way to resolve that? EDIT: updated my previous comment with an updated build - needed to toggle on the shared lib |
Please specify the FLATBUFFERS_BUILD_SHAREDLIB option. https://github.com/google/flatbuffers/blob/v23.5.26/CMakeLists.txt#L28 |
yeah I did that and still seeing
even though the output of
|
Add the path to LDFLAGS, |
nice, working for me requires flatbuffers v23.5.26 git clone https://github.com/google/flatbuffers.git
cd flatbuffers
git checkout tags/v23.5.26
cmake -G "Unix Makefiles" \
-DCMAKE_BUILD_TYPE=Release \
-DFLATBUFFERS_BUILD_SHAREDLIB:BOOL=ON
make
sudo make install
# if libedgetpu build fails
export LDFLAGS="-L/usr/local/lib" and hopefully we can get this merged |
Modified README with build instructions targeting TF v2.16.1 and FlatBuffers v23.5.26. |
new_makefile.patch.txt |
Still refers to ancient versions of TF. In reference to: google-coral#66 google-coral#72