Pimp My Llama - How to Customize LLMs for Your Business Case
summary
An off‑the‑shelf LLM becomes useful here by concrete steps: LLaMA 3 37B runs the Jonas simulator as a 24/7 Q&A brain, prompt engineering fixes tone and format cheaply, retrieval‑augmented generation pulls facts from 3,000 confluence docs to stop hallucinations, and fine‑tuning is possible but costly—full tuning needs racks of GPUs (individual datacenter GPUs ~€30,000) and lots of data—so RAG plus careful prompts usually delivers a working, auditable system without heavy tuning.
AI-written summary · about
chapters (8)
- 0:00Customizing LLMs introduction
- 5:20What are LLMs and types
- 8:51Hardware and costs locally
- 12:40Prompt engineering
- 21:31Retrieval Augmented Generation
- 37:24Fine-tuning LLMs
- 39:10Testing customized LLM
- 41:20Q&A
related talks
44mPimp My Llama - How to Customize LLMs for your Business Case
29mLocal LLM fine-tuning: a practical example
09mTailoring Llama for Your Needs: Fine-Tuning & Evaluation
29mCustomizing and Evaluating LLMs, an Ops Perspective
01h 49mAI engineering with open access LLMs that lie, curse, and steal
28mA community based open source LLM and what it can do for your business
49mTame your Llama, run AI in Java
31mCustomizing LLMs: A Guide to local Fine-Tuning...