Dear Allwinner Technical Support Team,
We are working on deploying an AI inference pipeline on the Cubie A7Z board which uses the Allwinner A733 SoC.
Board Details:
- Board: Cubie A7Z (A733 SoC)
- OS: Debian 11 Bullseye
- Kernel: 5.15.147-14-a733
- GLIBC Version: 2.31
What We Have Done:
- Successfully converted YOLOX model to .nb format using Pegasus/Acuity Toolkit (unified-tina AI-SDK v2.0.10.1)
- Successfully compiled inference binary natively on the board
- Confirmed NPU kernel driver (vipcore) is loaded and /dev/vipcore exists
- Confirmed NPU interrupt is registered at /proc/interrupts
The Problem:
All three variants of libGAL.so provided in the unified-tina AI-SDK require GLIBC 2.33:
- lib/glibc-gcc10_2_0/libGAL.so → requires GLIBC 2.33
- lib/glibc-gcc13_2_0/libGAL.so → requires GLIBC 2.33
- lib/aarch64-none-linux-gnu/libGAL.so → requires GLIBC 2.33
However the official Debian 11 Bullseye OS for Cubie A7Z only provides GLIBC 2.31. This causes the following error at runtime:
GLIBC_2.33 not found (required by libGAL.so)
Our Request:
Could you please provide:
- NPU userspace libraries (libGAL.so and all dependent .so files) compiled against GLIBC 2.31 for Debian 11 Bullseye
- OR the official Tina Linux image for Cubie A7Z / A733 SoC with pre-matched NPU libraries
This is blocking our production AI deployment on the Cubie A7Z board. Any urgent assistance would be greatly appreciated.