Welcome to MLE documentation!

MLE is a personal knowledge base for machine learning development.

Contents

• Python
  • Docstrings
  • Decorators
  • Hooks
  • Typing hints
  • Logging
  • Virtual environments
  • Debugging
  • Testing
  • Packaging
• Makefile
  • Manage ML experiments
• PyTorch
  • Model visualization
  • torch.compile
  • PyTorch profiling
    • Use PyTorch Profile
    • Use NVIDIA Nsight Systems
  • Use PyTorch Lightning
• CUDA
  • GPU architecture
  • Common CUDA optimizations
  • Hardware-aware algorithms
  • CUDA extension for PyTorch
  • References
• OpenAI Triton
  • Design rationales
  • Tutorials
  • Examples
• Kubenetes
  • Docker
    • Install the latest Docker
    • Install nvidia-docker2
    • Setup proxy for docker
  • Kind
    • Install kind
    • Use kind to create a k8s cluster
    • Load local images to kind
    • Troubleshooting kind
  • Known issues
    • Docker configuration not working
    • Timeout when pulling images in kind
• Inference engine
  • TorchScript
  • Microsoft ONNX
  • NVIDIA TensorRT
• LLM serving
  • Milestone models
  • Main challenges
  • Popular optimizations
  • Inference engine
    • vLLM
    • SGLang
  • Best practices
• Diffusion serving
  • Milestone models
  • Main challenges
  • Popular optimizations
  • Inference engines
• LLM applications
  • LLM and its limitations
  • RAG
    • Fine-tuning vs. RAG
    • Measurements of RAG
    • Challenges for RAG
    • Recent advancements
  • Agentic AI
  • Recipe for AI assistant
  • Recipe for AI scientist
  • References
• AIGC model
  • Modeling techniques for AIGC
  • Foundation model development
  • References