AIoT-MLSys-Lab/Efficient-LLMs-Survey

Provides a comprehensive taxonomy organizing efficient LLM techniques across model-centric (compression, architecture optimization, inference acceleration), data-centric (selection, curation), and framework-centric perspectives. Covers specific methods including quantization, pruning, low-rank adaptation, parameter-efficient fine-tuning (LoRA, adapters), speculative decoding, KV-cache optimization, and efficient attention mechanisms like grouped-query attention. Actively maintained repository with curated paper collection and taxonomy designed to be updated with emerging research.

1,256 stars. No commits in the last 6 months.