Quantization in Neural Networks - Basics Explained | Affine and Symmetric Quantization

Рет қаралды 2,365

Жыл бұрын

This tutorial explains the basics behind different quantization approaches explaining the math and the intuitions. Explains how the mapping is done from float32 precision to int8 precision.
Link to the Slides : drive.google.com/file/d/10kzMLVAVntXAr-lbOXc4RmGmX9iqjjIZ/view?usp=sharing
----------------------------------------------------------------------------------------------------------------
Reference materials for further reading.
A White Paper on Neural Network Quantization - arxiv.org/abs/2106.08295
Introduction to Quantization on PyTorch - pytorch.org/blog/introduction-to-quantization-on-pytorch/
Nvidia docs on Quantisation Basics - docs.nvidia.com/deeplearning/tensorrt/tensorflow-quantization-toolkit/docs/docs/intro_to_quantization.html
----------------------------------------------------------------------------------------------------------------
BGM Credits
🔻
Song: "Sappheiros - Falling (Ft. eSoreni) [Chill]" is under a Creative Commons license (CC-BY)
Music promoted by BreakingCopyright: bit.ly/Sappheiros-Falling
🔺

Пікірлер: 9

@heteromodal 5 ай бұрын

very helpful explanation, thank you!

@mittus1934 Жыл бұрын

Thanks for the content...Great work!

@user-kz5jc4tf7y Жыл бұрын

Great work Dingu

@dingusagar Жыл бұрын

Thanks Bhanu😊

@jeffr_ac Жыл бұрын

Cool! If you increased font size and showed actual use I think that'd add a lot of visibility to this video. The content and explanations are great.

@dingusagar Жыл бұрын

Thanks for the feedback. Let me check if I can do something about the font. Will create a follow-up video showing the practical use.

@dingusagar Жыл бұрын

Will fix the font in future videos. For now, I have added a link to the slides in the description. you can zoom into the slides if something is not legible.

@TheP0dcaster 11 ай бұрын

I would prefer without the background music 🥲

@dingusagar 11 ай бұрын

Point noted. thanks for the feedback. I put it to make it more engaging since it was a theory explanation. If it's not really helping and is a distraction, will remove it from next time onwards.