weight-compression

Neural Weight Compression: lossless LLM weight compression decoded inside the CUDA matvec. BF16 31 % smaller and faster than cuBLAS; fp8 13 % smaller at native fp8 speed. Bit-identical weights.

Post-training weight compression for low-RAM machines: Q4/Q8 quantization, green-format repair, AVX2 CPU inference, optional CUDA. ~45% less RAM at ~99.9% quality.

Run DFloat11 losslessly compressed BF16 checkpoints on Apple Silicon with MLX and a Metal decode kernel; today that runs FLUX.1 on a 32 GB Mac with weights about 30% smaller

SAEP-SVD: Feature-preserving SVD distillation for High-Rank DiT models beyond the Eckart-Young theorem (Z-Image / HighDream).

A deterministic, random-access archive format and toolkit for compressing, storing, diffing and patching large language model weights.