Lowaltitudeflight / local-llm-tuning-guide

git clone https://github.com/Lowaltitudeflight/local-llm-tuning-guide.git

A decision library for running local LLMs on multi-GPU machines without NVLink: parallelism, communication, quantization, speculative decoding, KV cache. Every conclusion carries its evidence grade and reproducible conditions.