WarpState: 150M non-Transformer LM trained from scratch
A developer trained a 150M-parameter non-Transformer language model called WarpState from scratch on 300M tokens. The project is an experimental proof of concept, not a claim of beating Transformers.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Claude Code silently deletes local history older than 30 days
- Split sigmas speed up video reference in Stable Diffusion
- MiniMax tool turns anything into realistic human video
- MirroS' Code-as-World rewrites videos into executable MuJoCo programs
- PRAXIST research system boosts MLE-bench scores 44%