Lessons learned finetuning Mistral-7B on over 3500 pages of trading commentary

Table of Contents

Introduction
Method
Step 1: Gather training data
Step 2: Clean and parse data
Step 3: Train the LLM
Step 4: Evaluate, Rinse & Repeat
Conclusion

Introduction

In the swirling world of financial market analysis, where change is as unpredictable as the Mistral, I embarked on an ambitious journey to create a large language model (LLM) that was an expert in the technical analysis of stock and options trading. Drawing from what I had learned making GPT-4 trade options, I knew the next frontier was to add more technical analysis expertise via fine-tuning. And what a frontier wrestling with a 7 billion (that’s billion with a B) parameter model was. The results of this rollercoaster though I am very excited to share with you all.