SPS Transformer: A New Dual-Stream Parallel Architecture
heghbalz · x · 2026-07-14
The shared post introduces a new Transformer architecture: SPS Transformer.
The core idea is to run two parallel streams for each input token—one starts from the input token and is responsible for building the state (kv cache), while the other directly predicts the next token via a <predict> token without writing to the cache. This design appears to be exploring more efficient inference/generation pathways.
Related event: SPS Transformer: Separating State and Prediction into Dual Streams(5 posts)→
More from Research
- Token Collapse: Study Finds Graph Soft Prompting Tokens in LLMs Fail to Encode Structure — Sapienza · 2026-10-27
- The 'sweet lesson' is real: two kinds of ML problems need opposite strategies — mike64_t · 2026-10-27
- PixGR: Pixel-space generative decoding for fast feed-forward 3D reconstruction — kwangmoo_yi · 2026-10-27
- UniAction: unified action conditioning lets one world model handle mixed action types — ZeYanjie · 2026-10-27
- Infini-AI Lab open-sources Hermit, a framework for agent harness evolution and continual learning — BeidiChen · 2026-10-27
- Meta FAIR's Self-Direction turns agent failures into reusable LoRA skills, +13.8pp on Terminal-Bench — karen_ullrich · 2026-10-27