High-Speed Inference with llama.cpp and Vicuna on CPU

For inference with large language models, we may think that we need a very big GPU or that it can’t run on consumer hardware. This is rarely the case. Nowadays, we have many tricks and frameworks at our disposal, such as device mapping or QLoRa, that make inference possible at home, even for very large language models. And now, thanks to Georgi Gerganov, we don’t even need a GPU. Georgi Gerganov is well-known for his work on implementing in plain C++ high-performance inference. He has implemented, with the help of many contributors, the inference for LLaMa, and other models, in plain C++. All these implementations are optimized to run without a GPU. In other words, you just need enough CPU RAM to load the models. Then your CPU will take care of the inference. In this blog post, I show how to set up a llama. cpp on your computer with very simple steps. I focus on Vicuna, a chat model behaving like ChatGPT, but I also show how to run llama.cpp for other language models. After reading this post, you should have a state-of-the-art chatbot running on your computer. <a href="https://medium.com/towards-artificial-intelligence/high-speed-inference-with-llama-cpp-and-vicuna-on-cpu-136d28e7887b">Read More</a>