InsightFace REST API for easy deployment of face recognition services with TensorRT in Docker.
👀 Apply YOLOv8 exported with ONNX or TensorRT(FP16, INT8) to the Real-time camera
Mixed-precision acceleration patch for MiniMax H3 inference on NVIDIA V100
Pytorch implementation of DreamerV2: Mastering Atari with Discrete World Models, based on the original implementation
ComfyUI custom node: run MiniMax H3 at near-fp16 speed with near-fp32 numerical stability on GPUs without bf16/fp8 hardware (V100 sm_70)
Let's train CIFAR 10 Pytorch with Half-Precision!
A flexible utility for converting tensor precision in PyTorch models and safetensors files, enabling efficient deployment across various platforms.
Export pytorch model to ONNX and convert ONNX from float32 to float 16
First super-resolution model designed for Apple Neural Engine. 2x upscale, real-time, on-device. Built by Ben Racicot.
apextrainer is an open source toolbox for fp16 trainer based on Detectron2 and Apex
40x faster AI inference: ONNX to TensorRT optimization with FP16/INT8 quantization, multi-GPU support, and deployment
5x faster, 3x smaller — TensorRT optimization pipeline for Meta SAM3 (40 FPS, split-module export, GPU preprocessing, mixed precision)
Transformer implementation in pytorch trained on NVIDIA A100 in fp16
Comprehensive performance analysis of DeepSeek V3 quantization levels (FP16, Q8_0, Q4_0) on 16GB GPU environments.
Backward pass CUDA kernels for fused GEMM+Bias+GeLU on SM75 (Turing). Float4 vectorized WMMA kernels validated against PyTorch autograd. 3.1x faster than autograd at M=1024. 24/24 tests passing.
Intelligent model optimization toolkit for edge and on-device AI, providing graph-aware quantization, mixed-precision optimization, compression, benchmarking, and hardware-aware deployment support.
A reproducible GPU benchmarking lab that compares FP16 vs FP32 training on MNIST using PyTorch, CuPy, and Nsight profiling tools. This project blends performance engineering with cinematic storytelling—featuring NVTX-tagged training loops, fused CuPy kernels, and a profiler-driven README that narrates the GPU’s inner workings frame by frame.
SuperPoint keypoint detector optimized for edge deployment: PyTorch → ONNX → TensorRT FP16. 32× speedup (14.66ms → 0.45ms, 2203 FPS) on Tesla T4. Targets Jetson Orin Ampere FP16 pipeline for real-time Visual SLAM.
To associate your repository with the fp16 topic, visit your repo's landing page and select "manage topics."