1. Prepare a supported board (We use Nexys A7-100T) and install required toolchains. See Setup Guide for more details
git clone https://github.com/scott306lr/AAML_Final.git
cd AAML_Final
./scripts/setup
# Automized script for building the project.
# Equivalent to "make build && make load", while fixing multiple definition of a non-constant variable.
# Enter the project directory
cd proj/AAML_final_proj
# Building project with the default model
bash run.sh
# Building project with our custom model
bash run.sh -m "model_compression/final_0.875_qat_model.tflite"
# Verbose mode, for debugging
bash run.sh -v
After the build process is finished, press enter and type 'reboot' to reboot and start the system.
After the program started, type 11g
to run test.
After the program started, press ctrl+c to leave litex-term, then run the following command:
python evaluation.py
Latency: 2sec -> 0.75sec (x2.67 speed up)