Quantization in Depth - Syllabus

• Introduction
  • Overview of advanced quantization techniques
  • Course objectives and key concepts
• Overview
  • Introduction to linear quantization
  • Benefits and challenges of model quantization
• Quantize and De-quantize a Tensor
  • Basic operations to quantize and de-quantize tensors
  • Code examples for tensor manipulation
• Get the Scale and Zero Point
  • Understanding scale and zero-point calculations
  • Practical examples and code implementation
• Symmetric vs Asymmetric Mode
  • Comparing symmetric and asymmetric quantization
  • Use cases for each mode
• Finer Granularity for More Precision
  • Exploring granularity options for higher precision
  • Code examples for fine-tuning quantization levels
• Per Channel Quantization
  • Implementing per-channel quantization
  • Benefits and trade-offs
• Per Group Quantization
  • Exploring per-group quantization techniques
  • Use cases and limitations
• Quantizing Weights & Activations for Inference
  • Quantization techniques for inference optimization
  • Practical examples of weight and activation quantization
• Custom Build an 8-Bit Quantizer
  • Building a custom 8-bit quantizer in PyTorch
  • Code examples for model layer replacement
• Replace PyTorch Layers with Quantized Layers
  • Techniques for replacing layers in PyTorch
  • Implementing quantized layers in existing models
• Quantize Any Open Source PyTorch Model
  • Guidelines for quantizing open source models
  • Using quantization for model efficiency
• Load Your Quantized Weights from Hugging Face Hub
  • Loading and using quantized weights from the Hugging Face Hub
  • Code examples for model deployment
• Weights Packing
  • Introduction to weight packing techniques
  • Benefits of packing for memory efficiency
• Packing 2-Bit Weights
  • Packing four 2-bit weights into one 8-bit integer
  • Code examples for weight packing
• Unpacking 2-Bit Weights
  • Unpacking techniques for 2-bit weight representations
  • Code examples for efficient model deployment
• Beyond Linear Quantization
  • Exploring advanced quantization techniques
  • Next steps for studying model compression
• Conclusion
  • Summary of quantization techniques and applications
  • Future directions for model optimization