This repository provides a simplebenchmarking suite for measuring the latency of a single forward pass through a HuggingFace Transformer model on CUDA GPUs. It supports both the standard PyTorch implementation and an optional Liger kernel acceleration for causal‑language models.