07-15-2020 05:21 AM - edited 07-15-2020 05:22 AM
Hi, @jasonwu
Xilinx has released the Vitis AI Library v1.2 recently. There, they have introduced Pytorch framework model development. For this framework, I couldn't find the VAI_C compiler and its usage.
I just want to know, whether this domain-specific compiler under development?
Thank you.
Natheesan
07-16-2020 07:36 AM
Hello,
please refer to UG1414, be sure this is the latest release 1.2 of 2020/07/07
vai_q_pytorch documentation starts from page 65
Compiler information at page 88:
For Pytorch, since the quantizer NNDCT will output the quantized model in XIR format directly, we will use vai_c_xir to compile it.
vai_c_xir -i /PATH/TO/quantized.xmodel -a /PATH/TO/arch/DPUCAHX8H/PLATFORM/arch.json -o /OUTPUTPATH -n netname
07-15-2020 06:12 AM
hello,
the Pytorch conda environment should be in GPU docker
07-15-2020 07:18 AM
Hi, @gguasti
Thanks for the response. Whether the next version of the Vitis AI User Guide (UG1414 V1.1) has been released?
Thank you.
Natheesan
07-15-2020 10:59 PM
Hi, @gguasti @jasonwu @jheaton @meherp
Regarding the Pytorch framework, I could only see some documentations (https://github.com/Xilinx/Vitis-AI/tree/master/Vitis-AI-Quantizer/vai_q_pytorch) related to Vitis AI Quantizer for Pytorch.
However, I couldn't see any documentation related to the Vitis AI Compiler for Pytorch (vai_c_pytorch). If you have any documents related to this compiler and its usage, could you share it with me?
Thanks,
Natheesan
07-16-2020 07:36 AM
Hello,
please refer to UG1414, be sure this is the latest release 1.2 of 2020/07/07
vai_q_pytorch documentation starts from page 65
Compiler information at page 88:
For Pytorch, since the quantizer NNDCT will output the quantized model in XIR format directly, we will use vai_c_xir to compile it.
vai_c_xir -i /PATH/TO/quantized.xmodel -a /PATH/TO/arch/DPUCAHX8H/PLATFORM/arch.json -o /OUTPUTPATH -n netname