From 2c943db9f8e69f5dd68161dce44a1c5f88ba605d Mon Sep 17 00:00:00 2001
From: Volod <141863857+volod-vana@users.noreply.github.com>
Date: Mon, 18 Mar 2024 12:58:46 -0400
Subject: [PATCH] feat: Docker support (#25)

Co-authored-by: Tim Nunamaker <tim@vana.com>
---
 .dockerignore |  1 +
 Dockerfile    | 53 +++++++++++++++++++++++++++++++++++++++++++++
 README.md     | 59 +++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 113 insertions(+)
 create mode 100644 .dockerignore
 create mode 100644 Dockerfile

diff --git a/.dockerignore b/.dockerignore
new file mode 100644
index 0000000..c04bc49
--- /dev/null
+++ b/.dockerignore
@@ -0,0 +1 @@
+poetry.lock
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..b28d447
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,53 @@
+# Base image for building the UI
+FROM node:18.19-alpine3.18 AS selfie-ui
+WORKDIR /selfie
+COPY selfie-ui/package.json selfie-ui/yarn.lock ./
+RUN yarn install --frozen-lockfile --non-interactive
+COPY selfie-ui/ .
+RUN yarn run build
+
+# Base image for the application
+FROM python:3.11 as selfie-base
+ENV PYTHONDONTWRITEBYTECODE 1
+ENV PYTHONUNBUFFERED 1
+ENV PIP_NO_CACHE_DIR=1
+WORKDIR /selfie
+COPY . .
+COPY --from=selfie-ui /selfie/out/ ./selfie-ui/out
+RUN pip install poetry --no-cache-dir
+RUN poetry config virtualenvs.create false
+RUN poetry install --no-interaction --no-ansi
+EXPOSE 8181
+
+# CPU-specific configuration
+FROM selfie-base as selfie-cpu
+CMD ["python", "-m", "selfie"]
+
+# GPU-specific configuration using an NVIDIA base image
+# Ensure that the chosen image version has wheels available at https://jllllll.github.io/llama-cpp-python-cuBLAS-wheels.
+# You can see what CUDA version an image uses at https://docs.nvidia.com/deeplearning/frameworks/support-matrix/index.html.
+# For example, the image nvcr.io/nvidia/pytorch:23.10-py3 uses CUDA 12.2, which is available at https://jllllll.github.io/llama-cpp-python-cuBLAS-wheels/AVX2/cu122.
+FROM nvcr.io/nvidia/pytorch:23.10-py3 as selfie-gpu
+COPY --from=selfie-base /selfie /selfie
+WORKDIR /selfie
+
+RUN pip install poetry --no-cache-dir
+ENV PATH="/root/.local/bin:$PATH"
+RUN poetry install --no-interaction --no-ansi
+
+RUN bash /selfie/scripts/llama-cpp-python-cublas.sh
+
+# --verbose and other options should probably be controlled by the user
+CMD ["poetry", "run", "python", "-m", "selfie", "--gpu", "--verbose"]
+
+# ARM64-specific configuration (Apple Silicon)
+FROM selfie-base as selfie-arm64
+# Uncomment below if additional dependencies are needed for ARM64
+# RUN apt-get update && apt-get install -y --no-install-recommends <arm64-specific-dependencies> && rm -rf /var/lib/apt/lists/*
+CMD if [ "$(uname -m)" = "arm64" ]; then \
+        echo "Running with GPU support"; \
+        python -m selfie --gpu; \
+    else \
+        echo "Running without GPU support"; \
+        python -m selfie; \
+    fi
diff --git a/README.md b/README.md
index 3ca2b27..d31daf4 100644
--- a/README.md
+++ b/README.md
@@ -127,6 +127,65 @@ For most users, the easiest way to install Selfie is to follow the [Quick Start]
 
 > **Note**: You can host selfie at a publicly-accessible URL with [ngrok](https://ngrok.com). Add your ngrok token (and optionally, ngrok domain) in `selfie/.env` and run `poetry run python -m selfie --share`.
 
+## Using Docker
+
+You can also run Selfie using Docker. To do so, follow these steps:
+
+1. Ensure that [Docker](https://www.docker.com) is installed.
+2. Clone or [download](https://github.com/vana-com/selfie) selfie repository.
+3. In a terminal, navigate to the project directory.
+4. Run the following commands for the image you want to use (CPU, GPU, or ARM64).
+
+This will start the server and the UI in your browser at http://localhost:8181.
+Your data will be stored in the `data` directory.
+This mounts your Hugging Face cache into the container, so you don't have to download the models again if you already
+have them.
+
+### CPU Image
+```bash
+docker build --target selfie-cpu -t selfie:cpu .
+
+docker run -p 8181:8181 \
+  --name selfie-cpu \
+  -v $(pwd)/data:/selfie/data \
+  -v $(pwd)/selfie:/selfie/selfie \
+  -v $(pwd)/selfie-ui:/selfie/selfie-ui \
+  -v $HOME/.cache/huggingface:/root/.cache/huggingface \
+  selfie:cpu
+```
+
+### Nvidia GPU Image
+```bash
+docker build --target selfie-gpu -t selfie:gpu .
+
+docker run -p 8181:8181 \
+  --name selfie-gpu \
+  -v $(pwd)/data:/selfie/data \
+  -v $(pwd)/selfie:/selfie/selfie \
+  -v $(pwd)/selfie-ui:/selfie/selfie-ui \
+  -v $HOME/.cache/huggingface:/root/.cache/huggingface \
+  --gpus all \
+  --ipc=host --ulimit memlock=-1 --ulimit stack=67108864 \
+  selfie:gpu
+````
+
+### MacOS ARM64 Image
+```bash
+DOCKER_BUILDKIT=0 docker build --target selfie-arm64 -t selfie:arm64 .
+
+docker run -p 8181:8181 \
+  --name selfie-arm64 \
+  -v $(pwd)/data:/selfie/data \
+  -v $(pwd)/selfie:/selfie/selfie \
+  -v $(pwd)/selfie-ui:/selfie/selfie-ui \
+  -v $HOME/.cache/huggingface:/root/.cache/huggingface \
+  selfie:arm64
+```
+
+> **Note**: on an M1/M2/M3 Mac, you may need to prefix the build command with `DOCKER_BUILDKIT=0` to disable BuildKit.
+> ```bash
+> DOCKER_BUILDKIT=0 docker build -t selfie:arm64 .
+> ```
 
 ## Setting Up Selfie