| """ |
| ROCm Forge — Deployment Templates |
| Dockerfile, bash scripts, and requirements.txt templates for ROCm deployment. |
| """ |
|
|
| |
| |
| |
| DOCKERFILE_TEMPLATE = """# ============================================================ |
| # ROCm Forge — Auto-Generated Dockerfile for AMD GPU Deployment |
| # Generated by ROCm Forge: The CUDA-to-AMD Migration Agent |
| # ============================================================ |
| |
| # Base image: ROCm {rocm_version} with Ubuntu 22.04 |
| FROM rocm/dev-ubuntu-22.04:{rocm_version}-complete |
| |
| # Set environment variables for ROCm |
| ENV ROCM_HOME=/opt/rocm |
| ENV PATH=$ROCM_HOME/bin:$ROCM_HOME/opencl/bin:$PATH |
| ENV LD_LIBRARY_PATH=$ROCM_HOME/lib:$ROCM_HOME/lib64:$LD_LIBRARY_PATH |
| ENV HIP_VISIBLE_DEVICES=0 |
| ENV HSA_OVERRIDE_GFX_VERSION={gfx_version} |
| ENV PYTORCH_ROCM_ARCH="gfx90a;gfx942" |
| |
| # Install system dependencies |
| RUN apt-get update && apt-get install -y \\ |
| python3 \\ |
| python3-pip \\ |
| python3-venv \\ |
| git \\ |
| wget \\ |
| && rm -rf /var/lib/apt/lists/* |
| |
| # Create working directory |
| WORKDIR /app |
| |
| # Install PyTorch for ROCm |
| RUN pip3 install --no-cache-dir \\ |
| torch --index-url https://download.pytorch.org/whl/rocm{rocm_version} |
| |
| # Install ROCm-compatible dependencies |
| {pip_install_commands} |
| |
| # Copy application code |
| COPY . /app/ |
| |
| # Verify GPU access |
| RUN python3 -c "import torch; print('PyTorch:', torch.__version__); print('HIP available:', torch.cuda.is_available())" |
| |
| # Default command |
| CMD ["python3", "{entry_point}"] |
| """ |
|
|
|
|
| |
| |
| |
| DOCKER_COMPOSE_TEMPLATE = """# ============================================================ |
| # ROCm Forge — Auto-Generated Docker Compose for AMD GPU |
| # ============================================================ |
| version: '3.8' |
| |
| services: |
| app: |
| build: . |
| devices: |
| - /dev/kfd:/dev/kfd |
| - /dev/dri:/dev/dri |
| group_add: |
| - video |
| - render |
| ipc: host |
| shm_size: '16gb' |
| environment: |
| - HIP_VISIBLE_DEVICES=0 |
| - HSA_OVERRIDE_GFX_VERSION={gfx_version} |
| - PYTORCH_ROCM_ARCH=gfx90a;gfx942 |
| volumes: |
| - ./data:/app/data |
| - ./models:/app/models |
| ports: |
| - "{port}:{port}" |
| """ |
|
|
|
|
| |
| |
| |
| DEPLOY_SCRIPT_TEMPLATE = """#!/usr/bin/env bash |
| # ============================================================ |
| # ROCm Forge — Auto-Generated Deployment Script |
| # Deploy AI workload on AMD Developer Cloud / ROCm |
| # ============================================================ |
| set -euo pipefail |
| |
| echo "==============================================" |
| echo " ROCm Forge — AMD GPU Deployment Script" |
| echo "==============================================" |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 1: Verify ROCm Installation |
| # ---------------------------------------------------------- |
| echo "[1/6] Verifying ROCm installation..." |
| if command -v rocminfo &> /dev/null; then |
| echo " ✅ ROCm is installed" |
| rocminfo | grep -E "Name:|Marketing" | head -4 |
| else |
| echo " ❌ ROCm not found. Install from: https://rocm.docs.amd.com/" |
| echo " Quick install: sudo apt install rocm-dev" |
| exit 1 |
| fi |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 2: Verify GPU Visibility |
| # ---------------------------------------------------------- |
| echo "[2/6] Checking GPU visibility..." |
| if command -v rocm-smi &> /dev/null; then |
| rocm-smi --showid --showproductname 2>/dev/null || echo " ⚠️ rocm-smi warning (non-critical)" |
| echo " ✅ GPU check complete" |
| else |
| echo " ⚠️ rocm-smi not found, skipping GPU check" |
| fi |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 3: Set Up Python Environment |
| # ---------------------------------------------------------- |
| echo "[3/6] Setting up Python environment..." |
| python3 -m venv rocm_env 2>/dev/null || python3 -m venv rocm_env |
| source rocm_env/bin/activate |
| pip install --upgrade pip setuptools wheel |
| echo " ✅ Virtual environment ready" |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 4: Install ROCm-Compatible PyTorch |
| # ---------------------------------------------------------- |
| echo "[4/6] Installing ROCm-compatible PyTorch..." |
| pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm{rocm_version} |
| echo " ✅ PyTorch (ROCm) installed" |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 5: Install Project Dependencies |
| # ---------------------------------------------------------- |
| echo "[5/6] Installing project dependencies..." |
| {pip_install_commands} |
| echo " ✅ Dependencies installed" |
| echo "" |
| |
| # ---------------------------------------------------------- |
| # Step 6: Validate Setup |
| # ---------------------------------------------------------- |
| echo "[6/6] Validating AMD GPU + PyTorch setup..." |
| python3 - << 'VALIDATION' |
| import torch |
| import sys |
| |
| print(f" Python: {{sys.version.split()[0]}}") |
| print(f" PyTorch: {{torch.__version__}}") |
| print(f" HIP available: {{torch.cuda.is_available()}}") |
| |
| if torch.cuda.is_available(): |
| for i in range(torch.cuda.device_count()): |
| print(f" GPU {{i}}: {{torch.cuda.get_device_name(i)}}") |
| |
| # Quick sanity test |
| x = torch.randn(1000, 1000, device='cuda') |
| y = torch.matmul(x, x.T) |
| print(f" Matrix multiply test: ✅ PASSED (result shape: {{y.shape}})") |
| print("") |
| print(" 🎉 AMD GPU setup is ready!") |
| else: |
| print(" ❌ No HIP GPU detected.") |
| print(" Check: HIP_VISIBLE_DEVICES, ROCm install, GPU permissions") |
| sys.exit(1) |
| VALIDATION |
| |
| echo "" |
| echo "==============================================" |
| echo " ✅ Deployment complete! Run your app with:" |
| echo " source rocm_env/bin/activate" |
| echo " python3 {entry_point}" |
| echo "==============================================" |
| """ |
|
|
|
|
| |
| |
| |
| REQUIREMENTS_TEMPLATE = """# ============================================================ |
| # ROCm Forge — Auto-Generated Requirements for AMD GPU |
| # Install PyTorch FIRST using: |
| # pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.2 |
| # Then install the rest: |
| # pip install -r requirements.txt |
| # ============================================================ |
| |
| {requirements} |
| """ |
|
|
|
|
| |
| |
| |
| AMD_CLOUD_GUIDE = """# ============================================================ |
| # ROCm Forge — AMD Developer Cloud Quick Start |
| # ============================================================ |
| |
| ## Getting Started with AMD Developer Cloud |
| |
| ### 1. Sign Up |
| - Visit: https://www.amd.com/en/developer/resources/developer-cloud.html |
| - Create an AMD Developer account |
| - Apply for AMD Developer Cloud credits |
| |
| ### 2. Launch a GPU Instance |
| - Select an AMD Instinct MI300X instance |
| - Choose Ubuntu 22.04 with ROCm pre-installed |
| - Instance will have ROCm, PyTorch, and essential tools ready |
| |
| ### 3. Connect to Your Instance |
| ```bash |
| ssh -i your-key.pem user@your-instance-ip |
| ``` |
| |
| ### 4. Verify GPU Access |
| ```bash |
| rocminfo |
| rocm-smi |
| python3 -c "import torch; print(torch.cuda.is_available())" |
| ``` |
| |
| ### 5. Run Your Migrated Code |
| ```bash |
| # Upload your code |
| scp -i your-key.pem -r ./your-project user@your-instance-ip:/home/user/ |
| |
| # Or clone from GitHub |
| git clone https://github.com/your-repo.git |
| |
| # Run the deployment script |
| chmod +x deploy_rocm.sh |
| ./deploy_rocm.sh |
| ``` |
| |
| ## Useful AMD Developer Cloud Commands |
| ```bash |
| rocm-smi # GPU status |
| rocm-smi --showmeminfo vram # VRAM usage |
| rocminfo # Device info |
| hipcc --version # HIP compiler version |
| rocprof # Profiling |
| ``` |
| """ |
|
|
|
|
| |
| |
| |
| ENV_SETUP_TEMPLATE = """#!/usr/bin/env bash |
| # ============================================================ |
| # ROCm Forge — Environment Variables for AMD GPU |
| # Source this file: source env_rocm.sh |
| # ============================================================ |
| |
| # ROCm paths |
| export ROCM_HOME=/opt/rocm |
| export ROCM_PATH=/opt/rocm |
| export PATH=$ROCM_HOME/bin:$PATH |
| export LD_LIBRARY_PATH=$ROCM_HOME/lib:$ROCM_HOME/lib64:$LD_LIBRARY_PATH |
| |
| # GPU visibility (adjust as needed) |
| export HIP_VISIBLE_DEVICES={gpu_ids} |
| |
| # Architecture target (MI300X) |
| export PYTORCH_ROCM_ARCH="gfx90a;gfx942" |
| export HSA_OVERRIDE_GFX_VERSION={gfx_version} |
| |
| # Performance tuning |
| export PYTORCH_HIP_ALLOC_CONF=expandable_segments:True |
| export HIP_FORCE_DEV_KERNARG=1 |
| export MIOPEN_FIND_MODE=3 |
| |
| # Memory |
| export GPU_MAX_HW_QUEUES=8 |
| |
| # Debugging (uncomment if needed) |
| # export HIP_LAUNCH_BLOCKING=1 |
| # export AMD_LOG_LEVEL=4 |
| # export MIOPEN_ENABLE_LOGGING=1 |
| |
| echo "ROCm environment configured." |
| echo "ROCm Home: $ROCM_HOME" |
| echo "HIP Devices: $HIP_VISIBLE_DEVICES" |
| """ |
|
|
