GPT-5.6 Sol: Discovering TensorRT Optimizations for RaCo-ALIKED-LightGlue+
GPT-5.6 Sol finds exact TensorRT graph optimizations for RaCo-ALIKED-LightGlue+, delivering a 2.67x median speedup over FP16 torch.compile while preserving n...
Recent posts
Float8 (FP8) Quantized LightGlue in TensorRT with NVIDIA Model Optimizer: up to ~6× faster and ~68% smaller engines
FP8 quantization via NVIDIA Model Optimizer shrinks TensorRT engines for SuperPoint + LightGlue and can cut latency versus FP32, with a visible match-quality...
Accelerating LightGlue Inference with ONNX Runtime and TensorRT
In this post, we show how to accelerate LightGlue inference using ONNX Runtime and TensorRT, achieving 2x-4x speed gains over compiled PyTorch, across many b...
