One Model to Loop Them All: Elastic Latent Recurrence for Language Models
We present Elastic Latent Recurrence (ELR), a recipe for training looped language models without this tax.
ProofPaper ↗
Key points
- A latent recurrent language model applies a shared block of layers repeatedly, so one checkpoint can trade inference compute for accuracy by changing the loop count.
- At one loop, an ELR model costs the same to serve as a non-recurrent model of the same size and, trained on the same data, reaches similar accuracy.
- Each additional loop then buys more accuracy for more inference compute.
- At inference, keeping only the last loop's attention KV cache holds attention KV memory at its one-loop level while retaining most of the gain from additional loops.
Sources (1)
- [1]One Model to Loop Them All: Elastic Latent Recurrence for Language ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 8, 06:05 PM
We present Elastic Latent Recurrence (ELR), a recipe for training looped language models without this tax.
A latent recurrent language model applies a shared block of layers repeatedly, so one checkpoint can trade inference compute for accuracy by changing the loop count.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 8, 2026The State of AI Report 2026
- Oct 8, 2026Bridging KV-Cache Quantization and Linear Attention: From Theory to Pretrained Weight Migration
- Oct 8, 2026MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement
- Oct 8, 2026One Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed Experts
- Oct 7, 2026ResidualQuant: KV Cache Quantization for Looped Transformers with 2-Bit Residuals
- Oct 7, 2026When Routing Reveals Membership: Privacy Leakage from MoE Router Telemetry