-
Notifications
You must be signed in to change notification settings - Fork 67
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
Add support for TF 2.16.1 #63
Conversation
@Namburger I know this doesn't have the visibility patch, yet, |
To conform with git tag
yeah, that sounds fine to me, as the current state of libedgetpu @ tf2.15 is already not building for libcoral and pycoral |
BTW, @Namburger is there any plan to provide actual binaries? while pycoral is distributed as a wheel (and so via Pypi, at least in principle), libedge is (according to the documentation anyway) distributed through either a debian repo, or a link for Windows or MacOS. Doesn't really affect me, but certainly those reading the documentation will expect to find updated binaries there, regardless on this github repo. Thanks. |
@feranick I don't have the infra to host libedgetpu packages to be installed via apt like previously, unfortunately. I think something we could do is to have a github CI package it on new commits to |
@Namburger I think using Github releases is a good strategy. In fact that is how MacOS/Windows binaries are currently served. In principle, changing the binaries in the zip files is all that would be requires (besides, changing the documentation accordingly). As for LInux packages, I have been making all sorts of deb packages that I hosted in my libedgetpu forked repo (even for the current release against TF 2.15), which you may be able to port directly into a new release for the official repo. https://github.com/feranick/libedgetpu/releases/tag/16.0TF2.16.1-1 I have x86_64 binaries for MacOS (I don't have M* macs) and I don't use Windows, so don't have binaries for those, but shouldn't be hard to make them. In any case, github may be indeed the easiest way to go, assuming the online documentation can be modified. |
(An irrelevant note of appreciation of what the edgetpu can still do despite its age. While being launched in 2019, with a rated 4 TOPS, it still perform quite well compared with highly hyped Desktop rated NPUs from Intel/AMD, supposedly reaching 10 TOPS. Google did something truly special, if not a bit underrated). |
yeah, I'm just as bump out about the state that the project is in :/ |
google-coral/edgetpu#841 |
This pull fails to maintain the makefile builds
|
This has worked for me quite well. Please provide full log and the system you are compiling for. |
I am running: git clone --depth 1 --branch "v2.16.1" https://github.com/tensorflow/tensorflow
git clone https://github.com/google-coral/libedgetpu
cd libedgetpu
export TFROOT=~/tensorflow
make -f makefile_build/Makefile -j$(nproc) libedgetpu seeing output:
@feranick probably somewhat related to #66 |
I can reproduce. I will take a look. In the meantime, the building with Docker/Bazel is recommended, as it works properly (see the README.md). (#66 is outdated and most already merged). |
I just pushed a new PR that adds support for TF 2.17.0, and allows for compilation of |
This PR builds libedgetpu against TF 2.16.1.