What’s in an LLM? Demystifying Hugging Face Models & How to Leverage Them For Business Impact

---

---

AI Summary

Webinar Summary

Introduction

• Speaker: Dr. Rebecca Bilbro, CTO and co-founder of Rotational Labs
• Company: Rotational Labs, a tech company that builds custom AI and language models for clients concerned with data privacy and intellectual property.
• Webinar Focus: Sharing best practices in applied machine learning and building a community of practice.

Speakers

• Patrick and Prima: Engineers at Rotational Labs, experts in applied machine learning.
• Current Project: Writing a book on event-driven machine learning for Manning Publications.
• Webinar Topic: Open source tools and techniques for building contextualized language models.

Large Language Models (LLMs)

• Popularity: ChatGPT gained 100 million users in 2 months, highlighting the business impact of LLMs.
• Challenges: Chatbots can provide incorrect or nonsensical responses, a phenomenon known as hallucination.
• Customization: Difficulties in customizing general LLMs for specific organizational needs.

Building Domain-Specific LLMs

• Advantages: Better results with proprietary data, control over data and model, cost savings, and avoiding vendor lock-in.
• Open Source Models: Hugging Face offers a repository of pre-trained models from major tech companies for various machine learning tasks.
• Hugging Face Resources: Inference endpoints, Enterprise Hub, access to GPUs, and machine learning expertise.

Training Domain-Specific LLMs

• Pre-training: Requires a lot of data and computational power, often not feasible for individual organizations.
• Fine-tuning: More common and feasible, involves training a model on a specific domain or task using a smaller dataset.

Use Cases Developed by Rotational Labs

• Supply Chain Risk Detection: Using sentiment analysis to reduce manual review of news articles.
• Questionnaire Evaluation: Fine-tuning an instruction model to grade questionnaire responses, avoiding third-party data sharing.

Community Engagement

• LinkedIn and YouTube: Rotational Labs shares updates and resources on these platforms.
• Twitch Streams: Hosted by Patrick, focusing on open source development.

Upcoming Events

• Next Webinar: Scheduled for September 24th, focusing on AI product management.

Q&A Highlights

• Data Requirements for Fine-Tuning: Starting with around a thousand instances may be sufficient.
• Using Models Without GPUs: Possible through quantization and efficient training methods.
• Open Source Models vs. Third-Party Tools: Open source models can be downloaded and used locally, maintaining data privacy.

Detailed Instructions and Tips

• No detailed instructions such as CLI commands, website URLs, or specific tips were provided in the transcript.