llama-cpp

Running a 27B Hybrid SSM Model with 128k Context on a 12 GB Laptop GPU

## The Problem Running a 27 billion parameter model with 128,000 token context on a laptop sounds like it needs a 48 GB GPU. Pure transformer models