Using LLaVA

LLaVA / BakLLaVA can be used with LLaMA.cpp.

You can find the full llama.cpp documentation here.

Step 1 - Clone the repo

git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp

Step 2 - Download the model

For example, we will use BakLLaVA-1 model, which is what is used on the demo instance.

Navigate to mys/ggml_bakllava-1 and download either q4 or q5 quant, as well as the mmproj-model-f16.gguf file.

The mmproj-model-f16.gguf file is necessary for the vision model.

Step 3 - Build the server

make server

Step 4 - Run the server

Read the llama.cpp documentation for more information on the server options. Or run ./server --help.

./server -t 4 -c 4096 -ngl 35 -b 512 --mlock -m models/openchat_3.5.Q5_K_M.gguf --mmproj models/mmproj-model-f16.gguf

Step 5 - Enable the server in the client

settings -> Vision -> Vision Backend -> LLaMA.cpp