| ## Using llama.cpp in the web UI | |
| #### Pre-converted models | |
| Simply place the model in the `models` folder, making sure that its name contains `ggml` somewhere and ends in `.bin`. | |
| #### Convert LLaMA yourself | |
| Follow the instructions in the llama.cpp README to generate the `ggml-model-q4_0.bin` file: https://github.com/ggerganov/llama.cpp#usage | |
| ## Performance | |
| This was the performance of llama-7b int4 on my i5-12400F: | |
| > Output generated in 33.07 seconds (6.05 tokens/s, 200 tokens, context 17) | |
| You can change the number of threads with `--threads N`. | |