Building LLMs from scratch - Data Batching in LLM instruction fine-tuning | Hands on project | Live Python coding

In this lecture, we learn everything about data batching in LLM instruction fine-tuning. We learn about the 5 steps involved in data batching: Format data using prompt template. Tokenize formatted data. Adjust to the same length with padding tokens. Create target token IDs for training. Replace padding tokens with placeholders. We also understand the importance of ignore_index = 100 in PyTorch and how to use it. This is a lecture explained through detailed whiteboard notes and live coding. The key reference book which this video series very closely follows is Build a Large Language Model from Scratch by Manning Publications. All schematics and their descriptions are borrowed from this incredible book! This book serves as a comprehensive guide to understanding and building large language models, covering key concepts, techniques, and implementations. Affiliate links for purchasing the book will be added soon. Stay tuned for updates! 0:00 Instruction finetuning recap so far 5:53 Batching the dataset introduction 9:08 Tokenizing formatted data 11:25 Padding token IDs 15:06 Creating target IDs for training 23:24 Replace padding tokens with “ignore index = -100” 24:31 Coding the Instruction Dataset class 27:50 Coding the custom collate padding function 32:04 Coding target token IDs 36:42 Coding the padding token replacement with “ignore index = -100” 41:07 Significance of PyTorch “ignore index = -100” 47:02 Masking target token IDs 50:40 Recap and summary Code file: https://drive.google.com/file/d/13jgqjSY-GOKhLoOsedUuZPXHTzQXrnt1/view?usp=sharing Instruction data link: https://github.com/rasbt/LLMs-from-scratch/blob/main/ch07/01_main-chapter-code/instruction-data.json Stanford Alpaca link: https://github.com/tatsu-lab/stanford_alpaca PyTorch cross entropy loss: https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html OpenAI Tiktoken library: https://github.com/openai/tiktoken Instruction Tuning with Loss Over Instructions paper: https://arxiv.org/abs/2405.14394 ================================================= ✉️ Join our FREE Newsletter: https://vizuara.ai/our-newsletter/ ================================================= Vizuara philosophy: As we learn AI/ML/DL the material, we will share thoughts on what is actually useful in industry and what has become irrelevant. We will also share a lot of information on which subject contains open areas of research. Interested students can also start their research journey there. Students who are confused or stuck in their ML journey, maybe courses and offline videos are not inspiring enough. What might inspire you is if you see someone else learning and implementing machine learning from scratch. No cost. No hidden charges. Pure old school teaching and learning. ================================================= 🌟 Meet Our Team: 🌟 🎓 Dr. Raj Dandekar (MIT PhD, IIT Madras department topper) 🔗 LinkedIn: https://www.linkedin.com/in/raj-abhijit-dandekar-67a33118a/ 🎓 Dr. Rajat Dandekar (Purdue PhD, IIT Madras department gold medalist) 🔗 LinkedIn: https://www.linkedin.com/in/rajat-dandekar-901324b1/ 🎓 Dr. Sreedath Panat (MIT PhD, IIT Madras department gold medalist) 🔗 LinkedIn: https://www.linkedin.com/in/sreedath-panat-8a03b69a/ 🎓 Sahil Pocker (Machine Learning Engineer at Vizuara) 🔗 LinkedIn: https://www.linkedin.com/in/sahil-p-a7a30a8b/ 🎓 Abhijeet Singh (Software Developer at Vizuara, GSOC 24, SOB 23) 🔗 LinkedIn: https://www.linkedin.com/in/abhijeet-singh-9a1881192/ 🎓 Sourav Jana (Software Developer at Vizuara) 🔗 LinkedIn: https://www.linkedin.com/in/souravjana131/