Installing the Factory SDK

This guide walks you through the process of installing the Factory SDK.

System Requirements

Before installing, ensure your system meets these requirements:

• Python: Version 3.10 or higher
• Operating System: Linux (x86_64 architectures)
• Hardware:
  • GPU: NVIDIA GPU with CUDA support (required)
  • Memory: At least 16GB RAM (32GB+ recommended for large-scale fine-tuning)
  • Disk Space: At least 10GB free for model weights and datasets

Basic Installation

Install the Factory SDK using pip:

pip install factory-sdk

Or if you use Poetry:

poetry add factory-sdk

This will install the latest available version of the SDK along with its dependencies like transformers, datasets, pydantic, and various ML libraries.

GPU Acceleration Setup

For proper functionality:

1. Ensure CUDA Toolkit is installed
2. Update your NVIDIA drivers
3. The SDK automatically includes acceleration libraries like bitsandbytes, deepspeed, and vllm

Verify your GPU setup with:

python -c "import torch; print(torch.cuda.is_available())"

This should return True if your GPU is properly configured.

Troubleshooting

Common Installation Issues

1. Package Conflicts

   • Issue: Dependency conflicts with existing packages
   • Solution: Use a virtual environment

     python -m venv factory-env
     source factory-env/bin/activate
     pip install factory-sdk

2. Permission Errors

   • Issue: "Permission denied" during installation
   • Solutions:

     pip install --user factory-sdk
     # Or update pip
     pip install --upgrade pip

3. Missing Dependencies

   • Issue: Installation fails due to missing libraries
   • Solution: Ensure you have all build dependencies

     # For Ubuntu/Debian
     sudo apt-get install python3-dev build-essential

4. GPU Not Detected

   • Issue: torch.cuda.is_available() returns False
   • Solutions:
     • Verify GPU is recognized: nvidia-smi
     • Check CUDA installation: nvcc --version
     • Reinstall CUDA if needed: sudo apt-get install --reinstall cuda

5. Version Compatibility

   • Issue: Package incompatible with Python version
   • Solution: Ensure you're using Python 3.10 or higher

     python --version
     # If needed, install correct version

6. Memory Errors During Import

   • Issue: Out of memory when loading large models
   • Solution: Increase swap space or use a system with more RAM