bfloat16 开源项目推荐（53 个）

libxsmm /libxsmm

用于专用稠密和稀疏矩阵运算以及深度学习基础原语的库。

VoidStarKat /half-rs

Half-precision floating point types f16 and bf16 for Rust.

JuliaMath /BFloat16s.jl

Julia implementation for the BFloat16 number type

Essenceia /Systolic_Array_with_DFT_v2

IHP 130nm ASIC tapeout of a 2x2 bfloat16 matrix matrix multiplication with DFT infrastructure. Iteration on the previous accelerator taped out on GF180.

Essenceia /BFloat16

RTL implementation of a performance/area optimized bfloat16 adder and multiplication.

higham /chop

Round matrix elements to lower precision in MATLAB

shibatch /tlfloat

C++ template library for floating point operations

DW0RKiN /Floating-point-Library-for-Z80

Floating-Point Arithmetic Library for Z80

KernelTuner /kernel_float

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

afterdusk /flop

IEEE 754-style floating-point converter

Essenceia /uselessly_fast_bfloat16_multiplier

Pushing the bf16 multiplication clock frequency to the max on the nominal corner on IHP 130nm 5L node.

aahouzi /llama2-chatbot-cpu

A LLaMA2-7b chatbot with memory running on CPU, and optimized using smooth quantization, 4-bit quantization or Intel® Extension For PyTorch with bfloat16.

nestordemeure /jochastic

A JAX implementation of stochastic addition.

sigurd4 /custom_float

Customizable floating point types, with all standard floating point operations implemented from scratch.

d4l3k /go-bfloat16

Bfloat16 conversion utilities for Go/Golang

Essenceia /Teapot

100Mbps Ethernet accelerator wrapper.

nestordemeure /stochastorch

A Pytorch implementation of stochastic addition.

StarOne01 /bfloat16

A lightweight C++ implementation of the Brain Floating Point (bfloat16) format.

sashakatne /nvfp4-dotprod-formal-dv

Formal (VC Formal FPV) and UVM verification of an 8-lane mixed-precision INT8/BF16/NVFP4 dot-product core, with a shared SystemVerilog golden reference across assertions and scoreboards.

imciner2 /ChopBLAS

Basic linear algebra routines implemented using the chop rounding function