Nightly builds available #330
tphakala
announced in
Announcements
Replies: 1 comment
-
can you comment on the performance gain? I'm seeing what looks to be really solid performance on a rpi5.
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I finally managed to create a nightly build workflow for BirdNET-Go, this means that if you do not wish to use Docker build you now have pre-release updates available at https://github.com/tphakala/birdnet-go/releases
Please note that this release contains an updated TensorFlow Lite library which allows use of the XNNPACK delegate feature for accelerated AI inference. Once you have downloaded the nightly build archive, copy the included library file to your library path replacing the existing libtensorflowlite_c.so. On Windows, delete the old library which had .so extension and keep the new DLL in the same directory as birdnet-go.exe.
Also note that current AMD64 releases requires AVX2 support from CPU, this is included on Intel CPUs since Haswell generation (4th Gen Intel Core, released in 2014).
Please backup your existing birdnet-go binary and tensorflow lite library in case you need to roll back
After you have verified that the updated BirdNET-Go and libtensorflow_c runs on your system, you can enable XNNPACK delegate by setting usexnnpack to true under the birdnet section in config.yaml.
Beta Was this translation helpful? Give feedback.
All reactions