model-compression-papers

A curated list of papers and code on efficient diffusion models for image, video, world modeling, and language generation. Covering acceleration, quantization, compression, caching, and distillation.

The official implementation of the ICML 2023 paper OFQ-ViT

Nexus ZLC is an experimental offline toolkit for lossless GGUF compression and compressed-weight inference. It preserves original model data, supports compatible models through a custom WebGPU engine, and includes restoration and before-and-after perplexity tests.

High-performance LLM compression engine using SVD matrix decomposition, INT8 quantization, and output caching training acceleration.

NEXUS LittleBit Compression & Recovery is an experimental AI platform supporting ultra-low-bit LLM quantization (8 to 0.1 BPW), custom AMD/WebGPU-accelerated inference, and five-stage recovery using up to 10 million training tokens. Designed to advance efficient, sovereign, offline AI on consumer hardware.

Papers for deep neural network compression and acceleration

From-scratch structured attention-head pruning — reproduces Michel et al. on GPT-2, shows where the method fails on a distilled model, and adds redundancy-scaling and layer-profile analyses.