
Activity
LLMs under control: constrained decoding with Guidance, Olive and ONNX Runtime
Online
About
A practical workshop on what happens when an application needs an LLM to return required fields, specific types and restricted values. It compares prompting, validation and constrained decoding.
What to expect
The session checks whether asking an LLM to answer in JSON is enough when an application needs required fields, specific types and restricted values.
- Model Phi with Microsoft Olive
- ONNX Runtime and JSON Schema
- Guidance/LLGuidance compared with prompting
- Build a reproducible structured generation pipeline
- Learn the limits of these techniques
FAQ
Where does it take place?
Online. The session has no physical venue and is held online.
How do I register?
Through the Meetup page:
https://www.meetup.com/microsoft-speakers-hub-en-espanol/events/317494925/
What tools are used?
The workshop works with:
A Phi model
Microsoft Olive
ONNX Runtime
JSON Schema
Guidance/LLGuidance
What will participants build?
A reproducible pipeline for structured generation. They will also compare prompting, validation and constrained decoding, and see the limits of these techniques.
Who organises it?
Microsoft Speakers Hub en Español.