vicuna / docs /llama.cpp-models.md
albertsoma's picture
Upload 123 files
16c358c
## 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`.