float8

PyTorch native quantization for training and inference

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

Official Code for the paper ELMO : Efficiency via Low-precision and Peak Memory Optimization in Large Output Spaces (in ICML 2025)

8-bit and sub-byte floating point types for Rust

A library written in C for converting between float8 (8-bit minifloat numbers) and float32 (single-precision floating-point numbers) formats.

minifloat (8-bit float) in Golang

Provide a dark-mode skin for Archive of Our Own inspired by Van Gogh’s The Starry Night, enhancing readability with a calming color scheme.

FP8 (E4M3) GEMM, GEMV and fused AdamW for NVIDIA GPUs without native FP8 (sm_50–sm_87, Maxwell to Ampere): exact products, FP32 accumulation, C API and PyTorch bindings incl. torch._scaled_mm and Float8Tensor

FP8 (E5M2 and E4M3, any pairing) GEMM, GEMV and fused AdamW for NVIDIA GPUs without native FP8 (sm_50–sm_87, Maxwell to Ampere): exact products, FP32 accumulation, TE HYBRID backward, C API and PyTorch bindings incl. torch._scaled_mm and Float8Tensor