Explaining How Real-Time GenAI Works in a Noisy Pub
"Almost everyone has heard about large language models, and tens of millions of people have tried out OpenAI ChatGPT and Google Bard. However, the intricate architecture and underlying mathematics driving these remarkable systems remain elusive to many. LLM's are fascinating - so let's grab a drink and find out how these systems are built and dive deep into their inner workings. In the length of time it to enjoy a round of drinks, you'll understand the inner workings of these models. We'll take our first sip of word vectors, enjoy the refreshing taste of the transformer, and drain a glass understanding how these models are trained on phenomenally large quantities of data. Large language models for your streaming application - explained with a little maths and a lot of pub stories"
Explaining How Real-Time GenAI Works in a Noisy Pub
- 1. Explaining how GenAI works ( in a noisy pub ) Simon Aubury @SimonAubury
- 2. GenAI You may have heard of it? 2
- 3. Training Pre training & 鍖ne tuning Large language models - 3 things Data A bunch of text Architecture The transformer 3
- 4. 1. Data Lets compress the internet
- 5. 5 Beer Image by Republica from Pixabay WebVectors Online for "beer"
- 6. Image by Christian_Birkholz from Pixabay
- 8. Image by delo from Pixabay
- 9. Understanding language can be challenging 9 The group of friends bought cold beer at the local pub to celebrate their reunion. context context
- 10. Transformers 10 Source: Attention Is All You Need https://arxiv.org/abs/1706.03762
- 12. A transformer + data = a base model 12 The base model Data Transformer Base model
- 13. Completion Base model 13 Completions Beer is served best cold with friends Context Base model + context = completion
- 14. 3. Training Pre training & 鍖ne-tuning
- 15. 15 How to train your LLM Pretraining (once) 1. PBs of text. 2. 1000s of GPUs. 3. Compress the text into a neural network, 4. Pay wait Obtain base model. Fine-tuning (recurring) 1. 1000s ideal Q&A responses (human) 2. Finetune base model on this data wait 3. Obtain assistant model 4. Evaluate, deploy & monitor
- 16. 鶲 It is faster to pick than to generate. 16 Reinforcement Learning from Human Feedback (RLHF) 鐃緒申 鐃緒申 鐃緒申
- 17. Training Pre training & 鍖ne tuning Large language models Data A bunch of text Architecture The transformer 17
- 18. 18 Retrieval Augmented Generation (RAG) What drink should I order?
- 19. The entire history of software engineering is one of rising levels of abstraction. Grady Booch, IBM chief scientist 19
- 20. 20 Thanks! @SimonAubury Presentation template by 際際滷sCarnival, Icons Flaticon photos Unsplash and Pixabay