Agent Zero AI framework
Added: Cuda Dockerfile. Edit: Initialize.sh to fix perms issue by adding -f force.
decib2d1aacd03e2
diff --git a/.gitignore b/.gitignoreindex c67214458..421e53075 100644--- a/.gitignore+++ b/.gitignore@@ -4,6 +4,12 @@**/__pycache__/**/.conda/+# Ignore docker/run/agent-zero directory+docker/run/agent-zero/++#Ignore cursor rules+.cursor/+# ignore test files in root dir/*.test.pydiff --git a/docker/run/Dockerfile.cuda b/docker/run/Dockerfile.cudanew file mode 100644index 000000000..85cc0ca64--- /dev/null+++ b/docker/run/Dockerfile.cuda@@ -0,0 +1,152 @@+# Use the NVIDIA CUDA base image with Ubuntu+FROM nvidia/cuda:12.8.1-base-ubuntu22.04++# Set non-interactive installation and timezone+ENV DEBIAN_FRONTEND=noninteractive+ENV TZ=UTC++# Check if the argument is provided, else throw an error+ARG BRANCH+RUN if [ -z "$BRANCH" ]; then echo "ERROR: BRANCH is not set!" >&2; exit 1; fi+ENV BRANCH=$BRANCH++# Set locale to en_US.UTF-8 and timezone to UTC (matching main Dockerfile)+RUN apt-get update && apt-get install -y locales tzdata+RUN sed -i -e 's/# \(en_US\.UTF-8 .*\)/\1/' /etc/locale.gen && \+ dpkg-reconfigure --frontend=noninteractive locales && \+ update-locale LANG=en_US.UTF-8 LANGUAGE=en_US:en LC_ALL=en_US.UTF-8+RUN ln -sf /usr/share/zoneinfo/UTC /etc/localtime+RUN echo "UTC" > /etc/timezone+RUN dpkg-reconfigure -f noninteractive tzdata+ENV LANG=en_US.UTF-8+ENV LANGUAGE=en_US:en+ENV LC_ALL=en_US.UTF-8+ENV TZ=UTC++# Copy contents of the project to root directory+COPY ./fs/ /++# Fix permissions for cron files from pre_install.sh+RUN chmod 0644 /etc/cron.d/*++# Install essential packages (from pre_install.sh but avoiding supervisor from apt)+RUN apt-get update && apt-get upgrade -y && apt-get install -y \+ python3 \+ python3-venv \+ python3-pip \+ nodejs \+ openssh-server \+ sudo \+ curl \+ wget \+ git \+ ffmpeg \+ cron++# Prepare SSH daemon (from pre_install.sh)+RUN bash /ins/setup_ssh.sh $BRANCH++# Configure Python 3.12 (specific to Ubuntu-based CUDA image)+RUN apt-get update && apt-get install -y \+ software-properties-common && \+ add-apt-repository -y ppa:deadsnakes/ppa && \+ apt-get update && \+ apt-get install -y python3.12 python3.12-venv python3.12-dev && \+ update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1 && \+ update-alternatives --set python3 /usr/bin/python3.12++# Bootstrap pip for Python 3.12 and install supervisor+RUN python3 -m ensurepip --upgrade && \+ python3 -m pip install --upgrade pip setuptools wheel && \+ python3 -m pip install supervisor++# Create supervisor directories+RUN mkdir -p /var/log/supervisor /etc/supervisor/conf.d++# Create root dotfiles with appropriate permissions+RUN touch /root/.bashrc /root/.profile && \+ chmod 644 /root/.bashrc /root/.profile++# Create a symlink for supervisord to the expected location+RUN ln -sf $(which supervisord) /usr/bin/supervisord++# Install additional software+RUN bash /ins/install_additional.sh $BRANCH++# Install core CUDA dependencies (minimized to essentials)+RUN apt-get update && apt-get install -y --no-install-recommends \+ cuda-cudart-dev-12-8 \+ libcublas-dev-12-8 \+ libcudnn8 \+ && apt-get clean \+ && rm -rf /var/lib/apt/lists/*++# Set CUDA environment variables+ENV PATH=/usr/local/cuda/bin:${PATH}+ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}+ENV CUDA_HOME=/usr/local/cuda+ENV CUDA_VERSION=12.8.1++# Install A0+RUN bash /ins/install_A0.sh $BRANCH++# Create and set up the shared instruments virtual environment+ENV INSTRUMENTS_VENV_PATH=/opt/instruments_venv+RUN python3 -m venv $INSTRUMENTS_VENV_PATH++# Switch to bash for the next RUN step (required for set -o pipefail)+SHELL ["/bin/bash", "-c"]++# Install all heavy dependencies into the instruments venv, with explicit checks+RUN set -euxo pipefail; \+ . $INSTRUMENTS_VENV_PATH/bin/activate; \+ echo "=== Upgrading pip, setuptools, wheel ==="; \+ pip install --upgrade pip setuptools wheel; \+ echo "=== Installing PyTorch with CUDA ==="; \+ pip install --no-cache-dir torch==2.6.0+cu124 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124; \+ echo "=== Installing other heavy dependencies ==="; \+ pip install --no-cache-dir \+ huggingface-hub==0.20.3 \+ safetensors==0.4.1 \+ accelerate==0.21.0 \+ diffusers==0.25.0 \+ transformers==4.38.2 \+ scipy==1.15.2 \+ xformers==0.0.29.post3; \+ echo "=== Checking all critical imports ==="; \+ python -c "import torch; print(f'PyTorch {torch.__version__} imported successfully. CUDA: {getattr(torch, 'cuda', None) and torch.cuda.is_available()}')" || (echo 'PyTorch import failed!' && exit 1); \+ python -c "import torchvision; print(f'Torchvision {torchvision.__version__} imported successfully.')" || (echo 'Torchvision import failed!' && exit 1); \+ python -c "import torchaudio; print(f'Torchaudio {torchaudio.__version__} imported successfully.')" || (echo 'Torchaudio import failed!' && exit 1); \+ python -c "import diffusers; print(f'Diffusers {diffusers.__version__} imported successfully.')" || (echo 'Diffusers import failed!' && exit 1); \+ python -c "import transformers; print(f'Transformers {transformers.__version__} imported successfully.')" || (echo 'Transformers import failed!' && exit 1); \+ python -c "import xformers; print(f'Xformers {xformers.__version__} imported successfully.')" || (echo 'Xformers import failed!' && exit 1); \+ python -c "import accelerate; print(f'Accelerate {accelerate.__version__} imported successfully.')" || (echo 'Accelerate import failed!' && exit 1); \+ python -c "import safetensors; print(f'Safetensors {safetensors.__version__} imported successfully.')" || (echo 'Safetensors import failed!' && exit 1); \+ python -c "import scipy; print(f'Scipy {scipy.__version__} imported successfully.')" || (echo 'Scipy import failed!' && exit 1); \+ python -c "import huggingface_hub; print(f'Huggingface_hub {huggingface_hub.__version__} imported successfully.')" || (echo 'Huggingface_hub import failed!' && exit 1); \+ echo "=== All critical imports succeeded ==="++# Optionally revert to sh for subsequent steps if needed+SHELL ["/bin/sh", "-c"]++# The existing global PyTorch install for /opt/venv (A0's main venv) is below.+# It might be used by other core A0 components, so we keep it.+# Our instruments will use the dedicated $INSTRUMENTS_VENV_PATH.+RUN . /opt/venv/bin/activate && \+ pip uninstall -y torch torchvision && \+ pip install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cu124++# Cleanup repo and install A0 without caching, this speeds up builds+ARG CACHE_DATE=none+RUN echo "cache buster $CACHE_DATE" && bash /ins/install_A02.sh $BRANCH++# Post installation steps+RUN bash /ins/post_install.sh $BRANCH++# Expose ports+EXPOSE 22 80++RUN chmod +x /exe/initialize.sh /exe/run_A0.sh /exe/run_searxng.sh++# Initialize runtime+CMD ["/exe/initialize.sh", "$BRANCH"]\ No newline at end of filediff --git a/docker/run/README.cuda.md b/docker/run/README.cuda.mdnew file mode 100644index 000000000..08359307b--- /dev/null+++ b/docker/run/README.cuda.md@@ -0,0 +1,104 @@+# Agent Zero: CUDA GPU Support 🚀++This guide explains how to build and run Agent Zero with NVIDIA GPU acceleration using CUDA. Running with CUDA enables faster performance for AI workloads by leveraging your GPU.++---++## Prerequisites++Before you begin, ensure you have:++1. **NVIDIA GPU** with CUDA capability+2. **NVIDIA Driver** installed on your host system+3. **NVIDIA Container Toolkit** ([Install Guide](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html))+ _This enables Docker to access your GPU_+4. **Docker** and **Docker Compose** installed++---++## 1. Build the CUDA Docker Image++Open a terminal in this directory and run:++```bash+# Set the branch you want to build from (default: main)+$branch="main"+docker build --no-cache -t frdel/agent-zero-run-cuda:testing --build-arg BRANCH=$branch -f Dockerfile.cuda .+```++---++## 2. Run Agent Zero with CUDA Support++You can start Agent Zero with GPU support using Docker Compose:++```bash+# On Linux, macOS, or Windows PowerShell:+docker-compose -f docker-compose.cuda.yml up -d+```++- This will launch Agent Zero in the background with GPU acceleration enabled.++---++## 3. Access Agent Zero++Once the container is running, open your browser and go to:++[http://localhost:50080](http://localhost:50080)++---++## 4. Stopping Agent Zero++To stop the CUDA-enabled Agent Zero container:++```bash+docker-compose -f docker-compose.cuda.yml down+```++---++## 5. Switching Between CPU and GPU Versions++You can easily switch between the CPU and GPU versions:++1. **Stop the currently running version:**+ ```bash+ # For CPU version:+ docker-compose down+ # For GPU version:+ docker-compose -f docker-compose.cuda.yml down+ ```++2. **Start the version you want:**+ ```bash+ # CPU version:+ docker-compose -f docker-compose.yml up -d++ # GPU (CUDA) version:+ docker-compose -f docker-compose.cuda.yml up -d+ ```++---++## Troubleshooting & Tips++- **First time setup may take several minutes** as dependencies are downloaded and installed.+- If you encounter issues with GPU access, verify your NVIDIA drivers and the NVIDIA Container Toolkit are correctly installed.+- To check if CUDA is available inside the container, you can run:+ ```bash+ docker exec -it <container_name> python3 -c "import torch; print(torch.cuda.is_available())"+ ```+- For advanced configuration, see the comments in [`Dockerfile.cuda`](mdc:docker/run/Dockerfile.cuda).++---++## More Information++- [NVIDIA Container Toolkit Documentation](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html)+- [Agent Zero Project](https://github.com/frdel/agent-zero) (replace with your actual repo link if different)++---++**Enjoy accelerated AI with Agent Zero and CUDA!**diff --git a/docker/run/docker-compose.cuda.yml b/docker/run/docker-compose.cuda.ymlnew file mode 100644index 000000000..2f3d7754b--- /dev/null+++ b/docker/run/docker-compose.cuda.yml@@ -0,0 +1,20 @@+services:+ agent-zero-cuda:+ container_name: agent-zero-cuda+ image: frdel/agent-zero-run-cuda:testing+ volumes:+ - ./agent-zero:/a0+ - ./agent-zero/work_dir:/root+ ports:+ - "50080:80"+ environment:+ - NVIDIA_VISIBLE_DEVICES=all+ - NVIDIA_DRIVER_CAPABILITIES=compute,utility+ - PYTHONUNBUFFERED=1+ deploy:+ resources:+ reservations:+ devices:+ - driver: nvidia+ count: all+ capabilities: [gpu]\ No newline at end of filediff --git a/docker/run/fs/exe/initialize.sh b/docker/run/fs/exe/initialize.shindex 170a162b2..438b2082b 100644--- a/docker/run/fs/exe/initialize.sh+++ b/docker/run/fs/exe/initialize.sh@@ -8,7 +8,7 @@ fiBRANCH="$1"# Copy all contents from persistent /per to root directory (/) without overwriting-cp -r --no-preserve=ownership,mode /per/* /+cp -r -f --no-preserve=ownership,mode /per/* /# allow execution of /root/.bashrc and /root/.profilechmod 444 /root/.bashrc