Dive deep into the realm of model quantization with the "Quantization in Depth" course, where you will learn to shrink model weights up to 4x their original size while maintaining performance. Whether you're a beginner or an enthusiast looking to enhance your understanding and practical skills in custom quantization techniques, this course led by experts from Hugging Face is your gateway to mastering advanced quantization methods.
What you will learn:
- Build and refine linear quantization functions, toggling between symmetric and asymmetric modes, and adapting to different granularities such as per-tensor, per-channel, and per-group.
- Assess and balance the trade-offs between performance and space through precise quantization error measurements.
- Develop a versatitle quantizer in PyTorch to apply quantization on any open-source model's dense layers, learning techniques to pack weights more efficiently.