Fine Tuning an LLM with the famous Alpaca Dataset and using Galileo to find errors
Using Galileo Fine-Tune you can improve the quality of your fine-tuned LLMs by improving the quality of your training data. Research has shown that small high-quality datasets can lead to powerful LLMs. Galileo Fine-Tune helps you achieve that.
Galileo integrates into your training workflow through its dataquality Python library. During Training, Galileo sees your samples and your model's output to find errors in your data. Galileo uses Guardrail Metrics as well as its Data Error Potential score to help you find your most problematic samples.
Data Error Potential (DEP) score
The Galileo Data Error Potential (DEP) score has been built to provide a per-sample holistic data quality score to identify samples in the dataset contributing to low or high model performance i.e. ‘pulling’ the model up or down respectively. In other words, the DEP score measures the potential for "misfit" of an observation to the given model.
Galileo surfaces token-level DEP scores to understand which parts of your Target Output or Ground Truth your model is struggling with.