TensorRT-LLM is NVIDIA's open-source library for optimizing Large Language Model inference on NVIDIA GPUs. It provides an easy-to-use Python API to define LLMs and applies state-of-the-art optimizations including custom attention kernels, inflight batching, paged KV caching, and quantization techniques such as FP8, FP4, INT4 AWQ, and INT8 SmoothQuant. Built on top of TensorRT and PyTorch, it delivers industry-leading throughput and latency for production LLM serving with support for MoE models and Blackwell GPU architectures.