mixed-precision 开源项目推荐（149 个）

stochasticai /xTuring

构建、个性化和控制您自己的LLM。从数据预处理到微调，xTuring提供了一种个性化开源LLM的简单方法。加入我们的不和谐社区：https://discord.gg/TgHXuSJEk6

NVIDIA /OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP

Azure /MS-AMP

Microsoft Automatic Mixed Precision Library

Zhen-Dong /HAWQ

Quantization library for PyTorch. Support low-precision and mixed-precision quantization, with hardware implementation through TVM.

mit-han-lab /haq

[CVPR 2019, Oral] HAQ: Hardware-Aware Automated Quantization with Mixed Precision

hellojialee /Improved-Body-Parts

Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation

thu-nics /ViDiT-Q

[ICLR'25] ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation

enp1s0 /ozIMMU

FP64 equivalent GEMM by the Ozaki scheme with Int8 Tensor Cores

moritztng /prism

High Resolution Style Transfer in PyTorch with Color Control and Mixed Precision 🎨

verificarlo /verificarlo

A tool for debugging and assessing floating point precision and reproducibility.

Tylogi /TyloQuant

Get more intelligence from every bit. Better quantization formats and smarter calibration let larger, stronger models run smoothly on the hardware you already own.

rickiepark /deep-learning-with-python-2nd

<케라스 창시자에게 배우는 딥러닝 2판> 도서의 코드 저장소

suvojit-0x55aa /mixed-precision-pytorch

Training with FP16 weights in PyTorch

tanyuqian /redco

NAACL '24（最佳演示论文亚军）/ MlSys @ NeurIPS '23 - RedCoast：一款自动化分布式训练与推理的轻量级工具

Zhen-Dong /BitPack

BitPack is a practical tool to efficiently save ultra-low precision/mixed-precision quantized models.

HuiResearch /tfbert

基于tensorflow1.x的预训练模型调用，支持单机多卡、梯度累积，XLA加速，混合精度。可灵活训练、验证、预测。

legritos /tfbert-xla-multigpu

TensorFlow 1.x BERT Training with Multi-GPU Gradient Accumulation XLA Mixed Precision 2026

andreped /GradientAccumulator

🎯 Gradient Accumulation for TensorFlow 2

EEESlab /CMix-NN

CMix-NN: Mixed Low-Precision CNN Library for Memory-Constrained Edge Devices

thu-nics /MixDQ

[ECCV24] MixDQ: Memory-Efficient Few-Step Text-to-Image Diffusion Models with Metric-Decoupled Mixed Precision Quantization