@samitouri / QOSami_Agents_Zero / commits / b2d1aacd

Added: Cuda Dockerfile. Edit: Initialize.sh to fix perms issue by adding -f force.

Added: Cuda Dockerfile. Edit: Initialize.sh to fix perms issue by adding -f force.

deci committed May 10, 2025 at 16:40 UTC b2d1aacd03e2f89acf958852811209f42cd9d802
5 files changed +283 -1
.gitignore
+6
@@ -4,6 +4,12 @@
4 **/__pycache__/
5 **/.conda/
6
7 +# Ignore docker/run/agent-zero directory
8 +docker/run/agent-zero/
9 +
10 +#Ignore cursor rules
11 +.cursor/
12 +
13 # ignore test files in root dir
14 /*.test.py
15
docker/run/Dockerfile.cuda new
+152
@@ -0,0 +1,152 @@
1 +# Use the NVIDIA CUDA base image with Ubuntu
2 +FROM nvidia/cuda:12.8.1-base-ubuntu22.04
3 +
4 +# Set non-interactive installation and timezone
5 +ENV DEBIAN_FRONTEND=noninteractive
6 +ENV TZ=UTC
7 +
8 +# Check if the argument is provided, else throw an error
9 +ARG BRANCH
10 +RUN if [ -z "$BRANCH" ]; then echo "ERROR: BRANCH is not set!" >&2; exit 1; fi
11 +ENV BRANCH=$BRANCH
12 +
13 +# Set locale to en_US.UTF-8 and timezone to UTC (matching main Dockerfile)
14 +RUN apt-get update && apt-get install -y locales tzdata
15 +RUN sed -i -e 's/# \(en_US\.UTF-8 .*\)/\1/' /etc/locale.gen && \
16 + dpkg-reconfigure --frontend=noninteractive locales && \
17 + update-locale LANG=en_US.UTF-8 LANGUAGE=en_US:en LC_ALL=en_US.UTF-8
18 +RUN ln -sf /usr/share/zoneinfo/UTC /etc/localtime
19 +RUN echo "UTC" > /etc/timezone
20 +RUN dpkg-reconfigure -f noninteractive tzdata
21 +ENV LANG=en_US.UTF-8
22 +ENV LANGUAGE=en_US:en
23 +ENV LC_ALL=en_US.UTF-8
24 +ENV TZ=UTC
25 +
26 +# Copy contents of the project to root directory
27 +COPY ./fs/ /
28 +
29 +# Fix permissions for cron files from pre_install.sh
30 +RUN chmod 0644 /etc/cron.d/*
31 +
32 +# Install essential packages (from pre_install.sh but avoiding supervisor from apt)
33 +RUN apt-get update && apt-get upgrade -y && apt-get install -y \
34 + python3 \
35 + python3-venv \
36 + python3-pip \
37 + nodejs \
38 + openssh-server \
39 + sudo \
40 + curl \
41 + wget \
42 + git \
43 + ffmpeg \
44 + cron
45 +
46 +# Prepare SSH daemon (from pre_install.sh)
47 +RUN bash /ins/setup_ssh.sh $BRANCH
48 +
49 +# Configure Python 3.12 (specific to Ubuntu-based CUDA image)
50 +RUN apt-get update && apt-get install -y \
51 + software-properties-common && \
52 + add-apt-repository -y ppa:deadsnakes/ppa && \
53 + apt-get update && \
54 + apt-get install -y python3.12 python3.12-venv python3.12-dev && \
55 + update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.12 1 && \
56 + update-alternatives --set python3 /usr/bin/python3.12
57 +
58 +# Bootstrap pip for Python 3.12 and install supervisor
59 +RUN python3 -m ensurepip --upgrade && \
60 + python3 -m pip install --upgrade pip setuptools wheel && \
61 + python3 -m pip install supervisor
62 +
63 +# Create supervisor directories
64 +RUN mkdir -p /var/log/supervisor /etc/supervisor/conf.d
65 +
66 +# Create root dotfiles with appropriate permissions
67 +RUN touch /root/.bashrc /root/.profile && \
68 + chmod 644 /root/.bashrc /root/.profile
69 +
70 +# Create a symlink for supervisord to the expected location
71 +RUN ln -sf $(which supervisord) /usr/bin/supervisord
72 +
73 +# Install additional software
74 +RUN bash /ins/install_additional.sh $BRANCH
75 +
76 +# Install core CUDA dependencies (minimized to essentials)
77 +RUN apt-get update && apt-get install -y --no-install-recommends \
78 + cuda-cudart-dev-12-8 \
79 + libcublas-dev-12-8 \
80 + libcudnn8 \
81 + && apt-get clean \
82 + && rm -rf /var/lib/apt/lists/*
83 +
84 +# Set CUDA environment variables
85 +ENV PATH=/usr/local/cuda/bin:${PATH}
86 +ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
87 +ENV CUDA_HOME=/usr/local/cuda
88 +ENV CUDA_VERSION=12.8.1
89 +
90 +# Install A0
91 +RUN bash /ins/install_A0.sh $BRANCH
92 +
93 +# Create and set up the shared instruments virtual environment
94 +ENV INSTRUMENTS_VENV_PATH=/opt/instruments_venv
95 +RUN python3 -m venv $INSTRUMENTS_VENV_PATH
96 +
97 +# Switch to bash for the next RUN step (required for set -o pipefail)
98 +SHELL ["/bin/bash", "-c"]
99 +
100 +# Install all heavy dependencies into the instruments venv, with explicit checks
101 +RUN set -euxo pipefail; \
102 + . $INSTRUMENTS_VENV_PATH/bin/activate; \
103 + echo "=== Upgrading pip, setuptools, wheel ==="; \
104 + pip install --upgrade pip setuptools wheel; \
105 + echo "=== Installing PyTorch with CUDA ==="; \
106 + pip install --no-cache-dir torch==2.6.0+cu124 torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124; \
107 + echo "=== Installing other heavy dependencies ==="; \
108 + pip install --no-cache-dir \
109 + huggingface-hub==0.20.3 \
110 + safetensors==0.4.1 \
111 + accelerate==0.21.0 \
112 + diffusers==0.25.0 \
113 + transformers==4.38.2 \
114 + scipy==1.15.2 \
115 + xformers==0.0.29.post3; \
116 + echo "=== Checking all critical imports ==="; \
117 + 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); \
118 + python -c "import torchvision; print(f'Torchvision {torchvision.__version__} imported successfully.')" || (echo 'Torchvision import failed!' && exit 1); \
119 + python -c "import torchaudio; print(f'Torchaudio {torchaudio.__version__} imported successfully.')" || (echo 'Torchaudio import failed!' && exit 1); \
120 + python -c "import diffusers; print(f'Diffusers {diffusers.__version__} imported successfully.')" || (echo 'Diffusers import failed!' && exit 1); \
121 + python -c "import transformers; print(f'Transformers {transformers.__version__} imported successfully.')" || (echo 'Transformers import failed!' && exit 1); \
122 + python -c "import xformers; print(f'Xformers {xformers.__version__} imported successfully.')" || (echo 'Xformers import failed!' && exit 1); \
123 + python -c "import accelerate; print(f'Accelerate {accelerate.__version__} imported successfully.')" || (echo 'Accelerate import failed!' && exit 1); \
124 + python -c "import safetensors; print(f'Safetensors {safetensors.__version__} imported successfully.')" || (echo 'Safetensors import failed!' && exit 1); \
125 + python -c "import scipy; print(f'Scipy {scipy.__version__} imported successfully.')" || (echo 'Scipy import failed!' && exit 1); \
126 + python -c "import huggingface_hub; print(f'Huggingface_hub {huggingface_hub.__version__} imported successfully.')" || (echo 'Huggingface_hub import failed!' && exit 1); \
127 + echo "=== All critical imports succeeded ==="
128 +
129 +# Optionally revert to sh for subsequent steps if needed
130 +SHELL ["/bin/sh", "-c"]
131 +
132 +# The existing global PyTorch install for /opt/venv (A0's main venv) is below.
133 +# It might be used by other core A0 components, so we keep it.
134 +# Our instruments will use the dedicated $INSTRUMENTS_VENV_PATH.
135 +RUN . /opt/venv/bin/activate && \
136 + pip uninstall -y torch torchvision && \
137 + pip install --no-cache-dir torch torchvision --index-url https://download.pytorch.org/whl/cu124
138 +
139 +# Cleanup repo and install A0 without caching, this speeds up builds
140 +ARG CACHE_DATE=none
141 +RUN echo "cache buster $CACHE_DATE" && bash /ins/install_A02.sh $BRANCH
142 +
143 +# Post installation steps
144 +RUN bash /ins/post_install.sh $BRANCH
145 +
146 +# Expose ports
147 +EXPOSE 22 80
148 +
149 +RUN chmod +x /exe/initialize.sh /exe/run_A0.sh /exe/run_searxng.sh
150 +
151 +# Initialize runtime
152 +CMD ["/exe/initialize.sh", "$BRANCH"]
\ No newline at end of file
docker/run/README.cuda.md new
+104
@@ -0,0 +1,104 @@
1 +# Agent Zero: CUDA GPU Support 🚀
2 +
3 +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.
4 +
5 +---
6 +
7 +## Prerequisites
8 +
9 +Before you begin, ensure you have:
10 +
11 +1. **NVIDIA GPU** with CUDA capability
12 +2. **NVIDIA Driver** installed on your host system
13 +3. **NVIDIA Container Toolkit** ([Install Guide](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html))
14 + _This enables Docker to access your GPU_
15 +4. **Docker** and **Docker Compose** installed
16 +
17 +---
18 +
19 +## 1. Build the CUDA Docker Image
20 +
21 +Open a terminal in this directory and run:
22 +
23 +```bash
24 +# Set the branch you want to build from (default: main)
25 +$branch="main"
26 +docker build --no-cache -t frdel/agent-zero-run-cuda:testing --build-arg BRANCH=$branch -f Dockerfile.cuda .
27 +```
28 +
29 +---
30 +
31 +## 2. Run Agent Zero with CUDA Support
32 +
33 +You can start Agent Zero with GPU support using Docker Compose:
34 +
35 +```bash
36 +# On Linux, macOS, or Windows PowerShell:
37 +docker-compose -f docker-compose.cuda.yml up -d
38 +```
39 +
40 +- This will launch Agent Zero in the background with GPU acceleration enabled.
41 +
42 +---
43 +
44 +## 3. Access Agent Zero
45 +
46 +Once the container is running, open your browser and go to:
47 +
48 +[http://localhost:50080](http://localhost:50080)
49 +
50 +---
51 +
52 +## 4. Stopping Agent Zero
53 +
54 +To stop the CUDA-enabled Agent Zero container:
55 +
56 +```bash
57 +docker-compose -f docker-compose.cuda.yml down
58 +```
59 +
60 +---
61 +
62 +## 5. Switching Between CPU and GPU Versions
63 +
64 +You can easily switch between the CPU and GPU versions:
65 +
66 +1. **Stop the currently running version:**
67 + ```bash
68 + # For CPU version:
69 + docker-compose down
70 + # For GPU version:
71 + docker-compose -f docker-compose.cuda.yml down
72 + ```
73 +
74 +2. **Start the version you want:**
75 + ```bash
76 + # CPU version:
77 + docker-compose -f docker-compose.yml up -d
78 +
79 + # GPU (CUDA) version:
80 + docker-compose -f docker-compose.cuda.yml up -d
81 + ```
82 +
83 +---
84 +
85 +## Troubleshooting & Tips
86 +
87 +- **First time setup may take several minutes** as dependencies are downloaded and installed.
88 +- If you encounter issues with GPU access, verify your NVIDIA drivers and the NVIDIA Container Toolkit are correctly installed.
89 +- To check if CUDA is available inside the container, you can run:
90 + ```bash
91 + docker exec -it <container_name> python3 -c "import torch; print(torch.cuda.is_available())"
92 + ```
93 +- For advanced configuration, see the comments in [`Dockerfile.cuda`](mdc:docker/run/Dockerfile.cuda).
94 +
95 +---
96 +
97 +## More Information
98 +
99 +- [NVIDIA Container Toolkit Documentation](https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html)
100 +- [Agent Zero Project](https://github.com/frdel/agent-zero) (replace with your actual repo link if different)
101 +
102 +---
103 +
104 +**Enjoy accelerated AI with Agent Zero and CUDA!**
docker/run/docker-compose.cuda.yml new
+20
@@ -0,0 +1,20 @@
1 +services:
2 + agent-zero-cuda:
3 + container_name: agent-zero-cuda
4 + image: frdel/agent-zero-run-cuda:testing
5 + volumes:
6 + - ./agent-zero:/a0
7 + - ./agent-zero/work_dir:/root
8 + ports:
9 + - "50080:80"
10 + environment:
11 + - NVIDIA_VISIBLE_DEVICES=all
12 + - NVIDIA_DRIVER_CAPABILITIES=compute,utility
13 + - PYTHONUNBUFFERED=1
14 + deploy:
15 + resources:
16 + reservations:
17 + devices:
18 + - driver: nvidia
19 + count: all
20 + capabilities: [gpu]
\ No newline at end of file
docker/run/fs/exe/initialize.sh
+1 -1
@@ -8,7 +8,7 @@ fi
8 BRANCH="$1"
9
10 # Copy all contents from persistent /per to root directory (/) without overwriting
11 -cp -r --no-preserve=ownership,mode /per/* /
11 +cp -r -f --no-preserve=ownership,mode /per/* /
12
13 # allow execution of /root/.bashrc and /root/.profile
14 chmod 444 /root/.bashrc