PROJECT / Open-source adaptation · upstream fork
KineJing-CausalWM
An attributed fork of AetherLabsAI/CausalWM with optional chunked VAE decoding and other engineering changes.
01 / Overview
KineJing-CausalWM
An attributed fork of AetherLabsAI/CausalWM with optional chunked VAE decoding and other engineering changes.
02 / Research question
Research question
How can memory pressure during frame decoding be reduced?
03 / Approach
Approach
Add optional chunked decoding and related checks to upstream code.
04 / Architecture and scope
Architecture and scope
Consult repository documentation and code for exact architecture and interfaces. This page describes only the public research scope.
05 / Current stage
Open-source adaptation
06 / Inspect the evidence
Inspect the evidence
View evidence ↗07 / Limitations
Limitations
Full-checkpoint inference, joint training and official scoring remain unverified; the upstream model is not original KineWorld work.
08 / CODE
Repository
https://github.com/kineworld/KineJing-CausalWM ↗09 / Next steps
Next steps
Follow new inspectable experiments and documentation in the repository.