CATs: Content-Addressable Transformers
summary
Claims an open-source framework that uses IPFS CIDs to content-address inputs processes outputs and IaC, orchestrates Spark workloads via Kubernetes (demo on Minikube), and records a Bill of Materials / invoice as a verifiable provenance chain so processes can be re‑executed. Demo builds a virtualenv, does spark-submit, writes CID'd partitions and a BOM, then replays a subsequent CAT on the network using an IPFS cache workaround; repo named cats targets Ubuntu 20.
AI-written summary · about
chapters (20)
- 0:00Introduction to CATs
- 1:48Why Content-Addressable Transformers are useful
- 3:36Decentralizing monoliths
- 5:06IPFS for verification
- 6:10Defining data products
- 7:54Data-driven design
- 9:00BOM as provenance
- 10:05CATs I/O pipeline
- 11:30Quantum principle
- 12:30Control planes
- 14:17Interoperable infrastructure
- 16:05CATs execution
- 17:07Kubernetes setup
- 18:34Executing CAT zero
- 19:32Demo: Deploy CAT
- 21:18Invoice and provenance
- 22:32Demo: Process data
- 24:18Generating results
- 26:22Acknowledgments
- 26:58Q&A
related talks
29mTokenization of Latent Computing Power
20mThe Future of the Web
13mLucid & Tiles — Simple, secure, and hackable IPFS dApps
24mFrom the Factory to Garage/Closet/Table: IPFS & Sustainability through Digital Twins
26mBuilding Apache Spark Clusters on K8s for AI Workloads
11mIPFS for Science
16mIPFS and Dapps: Obstacles and Opportunities
18mIPIP Protocol Improvements: Review & Working Session