LLM-Prompting-RAG-Aligment-Training-Tutorial

Introduction

During this tutorial session we will get to know how LLMs are trained. We will also explore creating our own LLM. You'll get a chance to prompt LLM, finetune an LLM, perform retrieval augmented generation, and implement LLM agents!

Slides available at: https://www.canva.com/design/DAGBJ2Rzgv8/ob5MPDkmOMoziqH22qtKoA/edit?utm_content=DAGBJ2Rzgv8&utm_campaign=designshare&utm_medium=link2&utm_source=sharebutton

The overview of the tutorial can be found below:

1 . Mistral 7B Prompting

Mistral 7B Prompting Introduction

Mistral 7B: https://arxiv.org/abs/2310.06825

Chain of Thought Prompting: https://arxiv.org/abs/2310.04959

Mistral 7B Code

https://github.com/MonikaVen/LLM-Prompting-RAG-Aligment-Training-Tutorial/blob/main/1-Mistral-7B-Effective-Prompting.ipynb

Load Mistral-7B model.
Prompt engineering for better results.
Metatags for better model performance.
Chain-of-thought prompting.

2. Mistral 7B RAG

Mistral 7B RAG Introduction

RAG: https://arxiv.org/abs/2005.11401

Mistral 7B RAG Code

https://github.com/MonikaVen/LLM-Prompting-RAG-Aligment-Training-Tutorial/blob/main/2-Mistral-7B-RAG.ipynb

Load Mistral-7B model.
Use custom data.
Use prompts that interact with your data.

3. Mistral 7B Align QLoRA

Mistral 7B Align Introduction

QLoRA: https://arxiv.org/abs/2305.14314

Mistral 7B Align Code

https://github.com/MonikaVen/LLM-Prompting-RAG-Aligment-Training-Tutorial/blob/main/3-Mistral-7B-Align-QLoRA.ipynb

Load Mistral-7B model.
Prepare dataset.
Use QLoRA to align the model.

4. OLMO framework LLM Training

OLMO Framework Introduction

OLMO: https://arxiv.org/abs/2402.00838

OLMO Framework Code

https://github.com/MonikaVen/LLM-Prompting-RAG-Aligment-Training-Tutorial/blob/main/4-OLMO-LLM-Training.ipynb

Get started with OLMO framework.
Load training data.
Create a training job.