NexaQuant: Llama.cpp-Compatible Model Compression with 100%+ Accuracy Recovery

Will llama.cpp be the go-to local inference framework for every device?