Data Preparation and Fine Tuning for Small language Models with AWS
29th September @ 12:00 pm - 4:30 pm
Not every job needs a large frontier model. A small model, fine-tuned on your own data, can outperform larger ones on specific tasks at a fraction of the cost and latency, whilst helping you to retain the edge your competitors can’t copy – your data.
This workshop will take you through implementing an end-to-end pipeline for fine-tuning a small LM (Llama 3.2 3B) to generate SQL queries from natural language using Amazon SageMaker’s serverless training and evaluation APIs. You’ll learn how to extract real user queries from an RDS PostgreSQL instance, clean and format them into training data; and then use the Notebook to orchestrates dataset registration, LoRA-based SFT training, and custom Lambda-based evaluation with MLflow experiment tracking.
Bring your laptop, and you’ll leave with a repeatable pipeline you can point at your own use case.
12.00pm Lunch & Welcome
12.30-12.40: Kick off and Overview
12.40-2.20: Workshop Part 1
2.20-2.30: Break
2.30-4.15: Workshop Part 2
4.15-4.30: Wrap up
Please note this event is targeted at startup builders and developers who already have some background knowledge of working with AI models. Attendance size will be strictly capped to ensure a quality experience for those who join.
Powered by Railpen.
Glasshouse Closed
From time to time, the Glasshouse will be fully closed to members to host private events. These may be due to space limitations, confidentiality, or resource constraints. Such events will be clearly marked as Glasshouse Closed.
Closed Meeting
Invite-only meetings or events where attendance is strictly limited to those personally invited. This may include confidential partner meetings or internal planning workshops. These will appear as Closed Meeting.
Open to Members
Events open to all members of the Glasshouse community. They are designed to encourage interaction, collaboration, and the exchange of ideas across sectors. Members are warmly encouraged to attend, connect, share, and grow their networks. These will appear as Open to Members.
Open to All
Events open to the wider innovation ecosystem, including Glasshouse members and friends. These gatherings are an opportunity to connect, exchange ideas, and collaborate across Cambridge and beyond. These will appear on the calendar as Open to All.