From 4fa73777716322d0990157b48b64db9cb8637842 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Iv=C3=A1n=20Baldo?= <ibaldo@netlabs.com.uy>
Date: Fri, 19 Jan 2024 16:50:20 -0300
Subject: [PATCH] Add ctranslate/Dockerfile with instructions to use it.

---
 ctranslate/Dockerfile | 39 +++++++++++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)
 create mode 100644 ctranslate/Dockerfile

diff --git a/ctranslate/Dockerfile b/ctranslate/Dockerfile
new file mode 100644
index 0000000..77b5750
--- /dev/null
+++ b/ctranslate/Dockerfile
@@ -0,0 +1,39 @@
+# From the parent directory (main directory of this repo) run:
+#
+# docker build --build-arg USERID=$(id -u) -t local/ctranslate-bench ctranslate
+#
+# 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/:Z \
+#   local/ctranslate-bench huggingface-cli login
+# Answer n to: Add token as git credential? (Y/n) n
+#
+# Needs 15G RAM free:
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z \
+#   -v$(pwd):/home/user/llama-inference:Z local/ctranslate-bench \
+#   sh -c 'cd /home/user/llama-inference/ctranslate && ct2-transformers-converter \
+#   --model meta-llama/Llama-2-7b-hf --output_dir llama-2-7b-ct2'
+#
+# docker run --rm -it -v$HOME/.cache/huggingface/:/home/user/.cache/huggingface/:Z \
+#   -v$(pwd):/home/user/llama-inference:Z --gpus all local/ctranslate-bench \
+#   sh -c 'cd /home/user/llama-inference/ctranslate && python3 bench.py'
+#
+# If using Podman with CDI substitute
+#   --gpus all
+# for
+#   --device nvidia.com/gpu=all --security-opt=label=disable
+
+# Select an available version from
+# https://github.com/OpenNMT/CTranslate2/pkgs/container/ctranslate2:
+FROM ghcr.io/opennmt/ctranslate2:latest-ubuntu20.04-cuda11.2
+
+RUN pip install --no-cache-dir torch transformers pandas
+
+# The parent image ENTRYPOINT does more harm than good, so disable it:
+ENTRYPOINT []
+
+ARG USERID=1000
+RUN adduser --disabled-password --gecos '' -u $USERID user
+USER user
+