low-precision

SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime

#学习与技能提升#Low Precision Arithmetic Simulation in PyTorch

A script to convert floating-point CNN models into generalized low-precision ShiftCNN representation

Low Precision(quantized) Yolov5

#大语言模型#JAX Scalify: end-to-end scaled arithmetics

Code for DNN feature map compression paper

CUDA/HIP header-only library for writing vectorized and low-precision (16 bit, 8 bit) GPU kernels

#自然语言处理#LinearCosine: Adding beats multiplying for lower-precision efficient cosine similarity