auto_gptq

auto_gptq

CONTAINERS IMAGES RUN BUILD

AutoGPTQ from https://github.com/PanQiWei/AutoGPTQ (installed under /opt/AutoGPTQ)

Inference Benchmark

Substitute the GPTQ model from HuggingFace Hub (or model path) that you want to run:

./run.sh --workdir=/opt/AutoGPTQ/examples/benchmark/ $(./autotag auto_gptq) \
   python3 generation_speed.py --model_name_or_path TheBloke/LLaMA-7b-GPTQ --use_safetensors --max_new_tokens=128

If you get the error Exllama kernel does not support query/key/value fusion with act-order, try adding --no_inject_fused_attention

CONTAINERS

auto_gptq
Builds
Requires	L4T >=34.1.0
Dependencies	build-essential cuda cudnn python tensorrt numpy cmake onnx pytorch torchvision huggingface_hub rust transformers
Dependants	l4t-text-generation text-generation-webui:1.7 text-generation-webui:6a7cd01 text-generation-webui:main
Dockerfile	Dockerfile
Images	dustynv/auto_gptq:r35.2.1 (2023-12-15, 6.0GB) dustynv/auto_gptq:r35.3.1 (2023-12-11, 6.0GB) dustynv/auto_gptq:r35.4.1 (2023-12-14, 6.0GB) dustynv/auto_gptq:r36.2.0 (2023-12-15, 7.7GB)

CONTAINER IMAGES

Repository/Tag	Date	Arch	Size
dustynv/auto_gptq:r35.2.1	2023-12-15	arm64	6.0GB
dustynv/auto_gptq:r35.3.1	2023-12-11	arm64	6.0GB
dustynv/auto_gptq:r35.4.1	2023-12-14	arm64	6.0GB
dustynv/auto_gptq:r36.2.0	2023-12-15	arm64	7.7GB

_{Container images are compatible with other minor versions of JetPack/L4T:
    • L4T R32.7 containers can run on other versions of L4T R32.7 (JetPack 4.6+)
    • L4T R35.x containers can run on other versions of L4T R35.x (JetPack 5.1+)}

RUN CONTAINER

To start the container, you can use the run.sh/autotag helpers or manually put together a docker run command:

# automatically pull or build a compatible container image
./run.sh $(./autotag auto_gptq)

# or explicitly specify one of the container images above
./run.sh dustynv/auto_gptq:r36.2.0

# or if using 'docker run' (specify image and mounts/ect)
sudo docker run --runtime nvidia -it --rm --network=host dustynv/auto_gptq:r36.2.0

^{run.sh forwards arguments to docker run with some defaults added (like --runtime nvidia, mounts a /data cache, and detects devices)
autotag finds a container image that's compatible with your version of JetPack/L4T - either locally, pulled from a registry, or by building it.}

To mount your own directories into the container, use the -v or --volume flags:

./run.sh -v /path/on/host:/path/in/container $(./autotag auto_gptq)

To launch the container running a command, as opposed to an interactive shell:

./run.sh $(./autotag auto_gptq) my_app --abc xyz

You can pass any options to run.sh that you would to docker run, and it'll print out the full command that it constructs before executing it.

BUILD CONTAINER

If you use autotag as shown above, it'll ask to build the container for you if needed. To manually build it, first do the system setup, then run:

./build.sh auto_gptq

The dependencies from above will be built into the container, and it'll be tested during. See ./build.sh --help for build options.