/setup-conda-rocm

You are helping the user set up a conda environment optimized for ROCm and PyTorch.

Process

1. Check if conda is installed

   • Run: conda --version
   • If not installed, suggest installing Miniconda or Anaconda
   • Installation: wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && bash Miniconda3-latest-Linux-x86_64.sh

2. Verify ROCm is available on system

   • Check: rocminfo
   • Get ROCm version: rocminfo | grep "Name:" | head -1
   • Typical ROCm versions: 5.7, 6.0, 6.1

3. Create conda environment

   conda create -n rocm-pytorch python=3.11 -y
   conda activate rocm-pytorch
   

4. Install PyTorch with ROCm support

   • Check compatible PyTorch version at: pytorch.org/get-started/locally/
   • Install based on ROCm version:

   # For ROCm 6.0
   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm6.0
   
   # For ROCm 5.7
   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/rocm5.7
   

5. Install essential ML libraries

   conda install -c conda-forge numpy scipy matplotlib jupyter ipython -y
   pip install pandas scikit-learn
   

6. Install deep learning tools

   pip install transformers accelerate datasets
   pip install tensorboard
   pip install onnx onnxruntime
   

7. Test PyTorch ROCm integration

   import torch
   print(f"PyTorch version: {torch.__version__}")
   print(f"CUDA available: {torch.cuda.is_available()}")  # ROCm uses CUDA API
   if torch.cuda.is_available():
       print(f"Device name: {torch.cuda.get_device_name(0)}")
       print(f"Device count: {torch.cuda.device_count()}")
   

8. Create activation script

   • Offer to create ~/scripts/activate-rocm-pytorch.sh:

     #!/bin/bash
     eval "$(conda shell.bash hook)"
     conda activate rocm-pytorch
     echo "ROCm PyTorch environment activated"
     python -c "import torch; print(f'PyTorch: {torch.__version__}, CUDA available: {torch.cuda.is_available()}')"
     

9. Optional: Install additional tools

   • Suggest:
     • timm - PyTorch image models
     • torchmetrics - Metrics
     • lightning - PyTorch Lightning
     • einops - Tensor operations

Output

Provide a summary showing:

• Conda environment name and Python version
• PyTorch version and ROCm compatibility
• GPU detection status
• List of installed packages
• Test results showing GPU is accessible
• Activation command for future use