fp16 开源项目推荐（74 个）

aiptimizer /TurboOCR

快速的 GPU OCR 服务器。FUNSD上的270 img/s。TensorRT FP16，PP-OCRv5，HTTP + gRPC。

SthPhoenix /InsightFace-REST

InsightFace REST API 便于在 Docker 中使用 TensorRT 轻松部署人脸识别服务。

Maratyszcza /FP16

Conversion to/from half-precision floating point formats

kamalkraj /stable-diffusion-tritonserver

Deploy stable diffusion model with onnx/tenorrt + tritonserver

petamoriken /float16

ES2025 float16 (IEEE 754 half-precision floating-point) ponyfill/proposal since 2017

the0807 /YOLOv8-ONNX-TensorRT

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

fengwang /float16_t

CPP20 implementation of a 16-bit floating-point type mimicking most of the IEEE 754 behavior. Single file and header-only.

Icbears /minimax-h3-v100-patch

Mixed-precision acceleration patch for MiniMax H3 inference on NVIDIA V100

higham /chop

Round matrix elements to lower precision in MATLAB

zerfoo /zerfoo

Pure Go machine learning framework. Train, run, and serve ML models with go build. Zero CGo.

esteveste /dreamerV2-pytorch

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

afterdusk /flop

IEEE 754-style floating-point converter

aaalll12322 /ComfyUI-MiniMaxH3-FP16Safe

ComfyUI custom node: run MiniMax H3 at near-fp16 speed with near-fp32 numerical stability on GPUs without bf16/fp8 hardware (V100 sm_70)

jizhuoran /caffe-android-opencl-fp16

Optimised Caffe with OpenCL supporting for less powerful devices such as mobile phones

kentaroy47 /pytorch-cifar10-fp16

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

MindAllik /GPU-Tensor-Unlock-Universal

Pre-driver FMLA32 Tensor Core unlocker and PTX benchmark console for GeForce RTX 40-series, RTX 30-series and RTX 20-series on Windows 10/11 (UEFI USB) and Linux (gpu_unlock.ko), with friendly UI for Win/Linux. Restores full 2x FP16/BF16/TF32->FP32 accumulate rate with automatic hardware profiling, pre-flight ROM checks, and zero VBIOS flashing.

angelolamonaca /PyTorch-Precision-Converter

Convert PyTorch checkpoints, safetensors files and whole Hugging Face model folders to fp16, bf16, fp8 or ComfyUI fp8_scaled, int8_convrot and int8, turn FP8 models (DeepSeek, Qwen) back into BF16 on a CPU, merge or split sharded safetensors, and extract EMA weights, with one command and without loading the model into RAM.

ModelPiper /PiperSR

First super-resolution model designed for Apple Neural Engine. 2x upscale, real-time, on-device. Built by Ben Racicot.

stlin256 /CUDABurner

An stress and benchmark utility for NVIDIA GPUs. Measures performance across various precisions (FP64, FP32, TF32, FP16, INT8) and monitors real-time vitals like power, temperature, and clock speeds.

waseemnabi08 /yolov8n-cnn-accelerator-fpga

Bachelor's Final Year Project an FPGA-based FP16 CNN accelerator for real-time YOLOv8n object detection on Xilinx Zynq UltraScale+ (ZCU104). Features a reconfigurable multi-engine architecture with on-chip dynamic retiling, achieving scalable throughput for edge AI applications.