A hobby project: a small Vision Transformer inference accelerator on the Basys 3 / Artix-7 FPGA (VHDL-2008) with a matching PyTorch QAT pipeline and integer golden model. FPGA output matches the reference on all 10,000 MNIST test images (77.21%).
fpga vhdl transformer mnist vivado basys3 artix-7 hardware-accelerator quantization-aware-training vision-transformer
-
Updated
Sep 16, 2026 - VHDL