Add a Dockerfile for the /hf benchmarks with instructions to build and run them.

hf/Dockerfile

@@ -0,0 +1,37 @@
+# From the parent directory (main directory of this repo) run:
+#
+# docker build --build-arg USERID=$(id -u) -t local/hf-bench hf
+#
+# If not already using and having a $HOME/.cache/huggingface/ then:
+#
+# mkdir $HOME/.cache/huggingface/
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/ local/hf-bench \
+#   huggingface-cli login
+# Answer n to: Add token as git credential? (Y/n) n
+#
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/ \
+#   -v$(pwd):/home/user/llama-inference --gpus all local/hf-bench \
+#   sh -c 'cd /home/user/llama-inference/hf && python3 bench.py'
+#
+# You can substitute bench.py by bench-bb.py, bench-gptq.py or any other.
+# If using Podman with CDI substitute
+#   --gpus all
+# for
+#   --device nvidia.com/gpu=all --security-opt=label=disable
+
+# Select an available version from
+# https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/supported-tags.md:
+# 2024-04-02 PyTorch was compiled for CUDNN8:
+#FROM nvcr.io/nvidia/cuda:12.3.2-cudnn9-runtime-rockylinux9
+FROM nvcr.io/nvidia/cuda:12.2.2-cudnn8-runtime-rockylinux9
+
+RUN dnf install -y \
+  python3-pip cuda-cupti-$(echo $CUDA_VERSION | sed -r 's/(.+)[.](.+)[.].*/\1-\2/') && \
+  dnf clean all && rm -rf /var/cache/dnf/*
+
+RUN pip install --no-cache-dir transformers accelerate optimum bitsandbytes auto_gptq scipy
+
+ARG USERID=1000
+RUN adduser -u $USERID user
+USER user
+