Arch Linux

paulbrause

Member

Registered: 2021-10-06

Posts: 40

[SOLVED] Is it possible to use a Google Coral M.2 Accelerator?

Hi,

i got a Google Coral M.2 Accelerator A+E key and i'm trying to install the drivers.
The Coral docs only mentions Debian packages, so i search in the AUR.

I found "gasket-dkms-git" and "libedgetpu-git", which should be the drivers and libraries etc.

The driver "gasket-dkms-git" installed without problems, but building "libedgetpu-git" always fails, and i don't know how to fix it.

I run:

yay -S libedgetpu-git

but i fails with:

.
.
.
Compiling tensorflow-2.18.0/tensorflow/lite/core/c/common.cc
Compiling tensorflow-2.18.0/tensorflow/lite/array.cc
In Datei, eingebunden von tensorflow-2.18.0/tensorflow/lite/schema/schema_generated.h:18,
                 von tensorflow-2.18.0/tensorflow/lite/util.cc:30:
tensorflow-2.18.0/tensorflow/compiler/mlir/lite/schema/schema_generated.h:26:41: Fehler: statische Assertion fehlgeschlagen: Non-compatible flatbuffers version included
   26 |               FLATBUFFERS_VERSION_MINOR == 3 &&
      |                                         ^
tensorflow-2.18.0/tensorflow/compiler/mlir/lite/schema/schema_generated.h:26:41: Anmerkung: der Vergleich wird zu »(12 == 3)« reduziert
Compiling /home/smart/.cache/yay/libedgetpu-git/src/libedgetpu/makefile_build/../tflite/edgetpu_context_direct.cc
Compiling /home/smart/.cache/yay/libedgetpu-git/src/libedgetpu/makefile_build/../tflite/edgetpu_manager_direct.cc
tensorflow-2.18.0/tensorflow/lite/util.cc: In Funktion »TfLiteStatus tflite::BytesRequired(TfLiteType, const int*, size_t, size_t*, TfLiteContext*)«:
tensorflow-2.18.0/tensorflow/lite/util.cc:200:21: Warnung: Vergleich von Ganzzahlausdrücken, von denen nur einer ein Vorzeichen hat: »int« und »size_t« {aka »long unsigned int«} [-Wsign-compare]
  200 |   for (int k = 0; k < dims_size; k++) {
      |                   ~~^~~~~~~~~~~
make: *** [makefile_build/Makefile:192: /home/smart/.cache/yay/libedgetpu-git/src/libedgetpu/makefile_build/../out/tensorflow-2.18.0/tensorflow/lite/util.o] Fehler 1
make: *** Es wird auf noch nicht beendete Prozesse gewartet …
==> FEHLER: Ein Fehler geschah in build().
    Breche ab...
 -> Fehler beim Erstellen: libedgetpu-git-exit status 4
 -> Die folgenden Pakete konnten nicht installiert werden. Ein manueller Eingriff ist erforderlich:
libedgetpu-git - exit status 4

Anyone an idea?
Thanks in advance...

Last edited by paulbrause (2025-01-08 01:20:25)

Lone_Wolf

Administrator

From: Netherlands, Europe

Registered: 2005-10-04

Posts: 15,038

Re: [SOLVED] Is it possible to use a Google Coral M.2 Accelerator?

It appears the package tries to build its own version of tensorflow .

Archlinux switched to python3.13 not long ago and tensorflow doesn't support python 3.13 yet you'll have to build against an older python.

arch-dev-public ML thread has more info .

---

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

paulbrause

Member

Registered: 2021-10-06

Posts: 40

Re: [SOLVED] Is it possible to use a Google Coral M.2 Accelerator?

I have installed Python versions 3.9.21, 3.10-dev (3.10.16+) and 3.11.11 with pyenv and started the installation again, but each time it aborts at the same point. Am I doing something wrong?

Lone_Wolf

Administrator

From: Netherlands, Europe

Registered: 2005-10-04

Posts: 15,038

Re: [SOLVED] Is it possible to use a Google Coral M.2 Accelerator?

tensorflow-2.18.0/tensorflow/compiler/mlir/lite/schema/schema_generated.h:26:41: Fehler: statische Assertion fehlgeschlagen: Non-compatible flatbuffers version included
   26 |               FLATBUFFERS_VERSION_MINOR == 3 &&
      |                                         ^
tensorflow-2.18.0/tensorflow/compiler/mlir/lite/schema/schema_generated.h:26:41: Anmerkung: der Vergleich wird zu »(12 == 3)« reduziert

maybe not.

archlinux flatbuffers version is 24.12.23 , build options can be found at https://gitlab.archlinux.org/archlinux/ … n/PKGBUILD

Try asking upstream for help solving this.

---

Disliking systemd intensely, but not satisfied with alternatives so focusing on taming systemd.

clean chroot building not flexible enough ?
Try clean chroot manager by graysky

paulbrause

Member

Registered: 2021-10-06

Posts: 40

Re: [SOLVED] Is it possible to use a Google Coral M.2 Accelerator?

Try asking upstream for help solving this.

I did and it seems, that there is a problem with the latest flatbuffers package. After downgrading to v24.3 the building process finishes as expected. Thank you!

Last edited by paulbrause (2025-01-08 01:19:57)