Discover ten essential tips for fine-tuning machine learning models in this informative video. Learn strategies like starting with small models, using LoRA or QLoRA, creating manual datasets, and implementing validation splits. Explore advanced techniques such as unsupervised fine-tuning and preference fine-tuning (ORPO). Gain insights on scaling up your training process, using logging tools, and applying these tips to multi-modal fine-tuning. Access additional resources including code repositories, advanced guides for vision, inference, and transcription, as well as support channels to enhance your fine-tuning skills.
Top Ten Tips for Fine-tuning Large Language Models
Overview
Syllabus
Top Ten Fine-tuning Tips
Tip 1: Start with a Small Model
Tip 2: Use LoRA or QLoRA
Tip 3: Create 10 manual questions
Tip 4: Create datasets manually
Tip 5: Start training with just 100 rows
Tip 6: Always create a validation data split
Tip 7: Start by only training on one GPU
Tip 8: Use weights and biases for logging
Scale up rows, tuning type, then model size
Tip 9: Consider unsupervised fine-tuning if you've lots of data
Tip 10: Use preference fine-tuning ORPO
Recap of the ten tips
Ten tips applied to multi-modal fine-tuning
Playlists to watch
Trelis repo overview
ADVANCED Fine-tuning repo Trelis.com/ADVANCED-fine-tuning
Training on completions only
ADVANCED fine-tuning repo CONTINUED
ADVANCED vision Trelis.com/ADVANCED-vision
ADVANCED inference trelis.com/enterprise-server-api-and-inference-guide/
ADVANCED transcription trelis.com/ADVANCED-transcription
Support + Resources Trelis.com/About
Taught by
Trelis Research
Related Courses
-
Fine-Tuning LLM Models - Generative AI Course
-
Improving Accuracy of LLM Applications
-
Full Fine-Tuning vs LoRA and QLoRA - Comparison and Best Practices
-
Master AI Efficiency with LoRA - Optimize Fine-Tuning for Large Language Models
-
Fine-Tuning a Diffusion Model with Your Photos
-
Creating Your Own Flux AI Model with LoRA Fine-Tuning
