Learning Fine-tuning

Learning Fine-tuning

LAMINI

Fine-tuning principles

Benefits

Tasks

Instruction finetuning

Definition

Data generation

can use open-source tools or chatgpt

Training process

for epoch in range(epochs):
    for batch in train_dataloader:
        outputs = model(**batch)
        loss = outputs.loss
        loss.backward()
        optimizer.step()

Evaluation

Introduction

Benchmarks

Error Analysis

Conclusion

Practical approach to finetuning

1. figure out your task
2. collect data related to the task’s inputs/outputs
3. Generate data if you don’t have enough data
4. finetune a small model (400m - 1b)
5. vary the amount of data you give the model
6. evaluate your LLM to know what’s going well vs not
7. collect more data to improve
8. increase task complexity
9. increase model size for performance