Recirculation Architecture Reduces Perplexity Across Gemma3 Models
xennygrimmato_ · x · 2026-08-20
Demonstrates the reduction in perplexity across the Gemma3 model series when using the Recirculation architecture. Comparisons show that the Recirculation transformer achieves lower perplexity than the Unrolled loop transformer.
Related event: Recirculation Architecture Cuts Perplexity Across Model Families(3 posts)→
More from Research
- Debunked: AI didn't invent a cancer vaccine overnight — it picks neoantigens — GaryMarcus · 2026-08-20
- Stanford Framework Boosts DeepSeek Past Claude at 1/11th Cost — FuSheng_0306 · 2026-08-20
- 3DGS-Uncertainty: Predictive Photometric Uncertainty for Gaussian Splatting — rsasaki0109 · 2026-08-20
- SemComp-Bench: Benchmarking Semantic Task Completion in Video Generation — FrameXAI · 2026-08-20
- Zetta ζ: Closed-Loop Embodied Harness Evolves Critics and Recovery Skills Online — Xin Ding · 2026-08-20
- SoftVTBench: A Deformation-Aware Visuo-Tactile Dataset for Deformable-Object Manipulation — TuojingAI · 2026-08-20