Transformer implementation in pytorch trained on NVIDIA A100 in fp16

A flexible utility for converting tensor precision in PyTorch models and safetensors files, enabling efficient deployment across various platforms.

apextrainer is an open source toolbox for fp16 trainer based on Detectron2 and Apex

Export pytorch model to ONNX and convert ONNX from float32 to float 16

Let's train CIFAR 10 Pytorch with Half-Precision!

Pytorch implementation of DreamerV2: Mastering Atari with Discrete World Models, based on the original implementation

👀 Apply YOLOv8 exported with ONNX or TensorRT(FP16, INT8) to the Real-time camera

InsightFace REST API for easy deployment of face recognition services with TensorRT in Docker.