Practical AI Creating instruction tuned models
8 snips
May 16, 2023 Erin Mikail Staples, a developer community advocate at Label Studio, shares her insights on creating instruction-tuned large language models. She emphasizes the importance of harnessing human feedback to refine AI outputs and demonstrates how context shapes model behavior. The discussion also highlights the critical role of open data and ethical labeling in enhancing machine learning accuracy. Erin's passion for improving AI accessibility and transparency makes this conversation a must-listen for anyone interested in custom generative models.
AI Snips
Chapters
Transcript
Episode notes
Bloomberg's Financial Model
- Bloomberg used their extensive financial data to retrain a large language model.
- This allowed them to create a model specifically tailored to financial information, improving data access and analysis.
Experiment with RLHF
- Explore an open-source reinforcement learning model built with GPT-2.
- Experiment with this model in a Google Colab notebook to understand how RLHF works.
Context Matters in Model Selection
- Consider the context of your problem before choosing a model.
- Smaller models are often sufficient, saving resources and preventing overfitting.