Context / Motivation
Inference traces can be extremely large. VizFold needs the ability to incrementally add new components (e.g., additional layers or representations) to an existing archive without recomputing everything from scratch.
Tasks
Design an API for incremental archive updates. Implement utilities to:
- Append new layers’ activations
- Add attention maps post hoc
- Validate archive integrity after updates
- Ensure compatibility with the chosen archive format
- Add unit tests for incremental updates
Expected Outcomes
- Utilities that support partial and incremental trace generation
- Reduced recomputation cost for experimentation
- Clear API for future model extensions
Skills Involved
- Data engineering
- ML systems
- Software architecture
- Testing and validation
Context / Motivation
Inference traces can be extremely large. VizFold needs the ability to incrementally add new components (e.g., additional layers or representations) to an existing archive without recomputing everything from scratch.
Tasks
Design an API for incremental archive updates. Implement utilities to:
Expected Outcomes
Skills Involved