| license: mit | |
| tags: | |
| - docker | |
| - llama.cpp | |
| - cuda | |
| - a100 | |
| # llama.cpp server-cuda (Docker image tar) | |
| Saved Docker image `ghcr.io/ggml-org/llama.cpp:server-cuda`, pulled and exported with `docker save` for offline deployment on CUDA 12 / A100 machines without internet access to GHCR. | |
| ## File | |
| - `llama.cpp-server-cuda.tar` (~2.6GB) — full image, load with `docker load`. | |
| ## Load and run | |
| ```bash | |
| docker load -i llama.cpp-server-cuda.tar | |
| docker run --rm --gpus all \ | |
| -v /path/to/models:/models \ | |
| -p 8080:8080 \ | |
| ghcr.io/ggml-org/llama.cpp:server-cuda \ | |
| -m /models/clef-flash-Q8_0.gguf \ | |
| --mmproj /models/clef-flash-vision.gguf \ | |
| -c 16384 -ngl 999 \ | |
| --host 0.0.0.0 --port 8080 | |
| ``` | |
| GGUF weights for clef-flash: https://huggingface.co/cuongpp/clef-flash-gguf | |