Xilinx / Vitis-AI-Tutorials

MIT License
358 stars 144 forks source link

Modify the repository please #92

Open AjinkyaDeshpande39 opened 1 year ago

AjinkyaDeshpande39 commented 1 year ago

The current Vitis-AI version is 2.5. In the https://github.com/Xilinx/Vitis-AI-Tutorials/tree/1.4/Introduction/02-System_Setup you say this whole tutorial is for vitis-ai v1.3. Then you ask people to pull the latest docker by

docker pull xilinx/vitis-ai:latest

It pulls v2.5. Or will pull the latest one available at the time you read this issue. The compiler in this version supports DPUCZDX8G_ISA1 series DPU architecture.

In the later stages, https://github.com/Xilinx/Vitis-AI-Tutorials/tree/1.4/Introduction/03-Basic/Module_1 you ask people to download pre built image of zcu104 which downloads vitis-ai v1.4. i.e. DPUCZDX8G_ISA0 series DPU architecture.

In later stages, https://github.com/Xilinx/Vitis-AI-Tutorials/tree/1.4/Introduction/03-Basic/Module_4 you will then have to edit arch.json file providing matching fingerprint. While compiling the model(provided you have changed the docker specific /opt/vitis_ai/compiler/arch/DPUCZDX8G/ZCU104/arch.json file), you will face error stating that the compiler for this architecture is not available. Furthermore, a proper documentation is also not available to know which version suits which DPU architecture. It took me a long time to figure out that this is actually docker's version problem. Later I pulled v1.4

docker pull xilinx/vitis-ai:1.4.916  

Then it compiled properly, target was prepared, copied to zcu104 and ran app_mt.py. finally it worked.

So i suggest to please update the repo.... Also, please mention there that arch.json file is specific to a particular version and can be accessed only through docker invoked terminal!

Thank you!