324-pytorch-to-vhdl

What problem does it solve?

Converts PyTorch neural networks into synthesizable VHDL so you can deploy bit-accurate inference on FPGA/ASIC hardware without manual RTL rewriting.

Core Features & Use Cases

• Fixed-point quantization workflow: Converts model weights to hardware-friendly integer formats such as Q8.8 and Q1.15.
• Layer-to-RTL mapping orchestration: Maps common PyTorch layers (e.g., nn.Linear, nn.ReLU, activations) to VHDL-friendly structures like MAC arrays and comparators using a layer rules knowledge base.
• Simulation-based verification: Generates and runs GHDL simulation artifacts to validate that RTL behavior matches quantized Python expectations.

Use case: You have a trained PyTorch classifier and need a deterministic, FPGA-optimized inference pipeline with verified fixed-point behavior.

Quick Start

Use the 324-pytorch-to-vhdl skill to convert your PyTorch model into synthesizable VHDL using Q8.8 quantization and generate a matching GHDL testbench for cycle-accurate verification.