Recurrent Looped Transformer adds per-token feedback to boost length generalization
The paper introduces the Recurrent Looped Transformer (RLT), which splits its eight layers between a parallel causal encoder and a recurrent decoder that merges the encoder output with the previous token's final decoder state, so the computation path grows with sequence length at a fixed per-token cost. On six algorithmic tasks, two RLT splits trained on at most 40 bits generalize parity to 256 bits with 100% accuracy in every seed while an eight-layer Transformer stays at chance; swap-based S5 permutation tracking at eight times the training length reaches 97% versus under 1%, and modular arithmetic reaches up to 93% versus 33%. Ablations show the gains depend on the feedback: removing it drops parity and S5 to chance, and updating feedback once per four-token chunk keeps 64-bit parity at 99% but lowers length-64 S5 from 100% to 20%.
Why it matters: It offers a simple architectural route to decoupling depth from sequence length, though the evidence so far comes only from small-scale synthetic algorithmic tasks.
Computer Science > Computation and Language
arXiv:2610.07591 (cs)
[Submitted on 6 Oct 2026]
Title:Recurrent Looped Transformer
Authors:Yifan Zhang, Jichen Feng, Shihan Qin
View a PDF of the paper titled Recurrent Looped Transformer, by Yifan Zhang and 2 other authors
View PDF HTML (experimental)Abstract:State tracking requires an update at every input, but the depth a Transformer applies to each token is fixed regardless of sequence length. We introduce the Recurrent Looped Transformer (RLT), which splits its layers between a parallel causal encoder and a recurrent decoder. At each token, the decoder merges the encoder output with the previous token's final decoder state, so the computation path grows with sequence length at a fixed per-token cost. On six algorithmic tasks, we compare five splits of eight layers with an eight-layer Transformer over three seeds. Trained on at most 40 bits, two RLT splits generalize parity to 256 bits with 100% accuracy in every seed, while the Transformer stays at chance. On swap-based $S_5$ permutation tracking at eight times the training length, RLT reaches 97% final-state accuracy versus under 1% for the Transformer, and accuracy increases with decoder depth. On modular arithmetic beyond the training lengths, RLT reaches up to 93% versus 33% for the Transformer. Ablations show that these gains depend on the feedback: removing it drops parity and swap-based $S_5$ to chance at every split. Updating the feedback once per four-token chunk lets known tokens in a chunk run in parallel and keeps 64-bit parity at 99%, while permutation tracking depends on per-token feedback: chunking lowers length-64 swap-based $S_5$ from 100% to 20%.
| Comments: | Project Page: this https URL |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.07591 [cs.CL] |
| (or arXiv:2610.07591v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.07591 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yifan Zhang [view email][v1] Tue, 6 Oct 2026 01:29:46 UTC (1,042 KB)
Access Paper:
- View PDF
- HTML (experimental)
- TeX Source
View a PDF of the paper titled Recurrent Looped Transformer, by Yifan Zhang and 2 other authors
Additional Features
Current browse context:
cs.CL
Change to browse by:
References & Citations
Loading...
BibTeX formatted citation
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, MediaCode, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
DemosDemos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related PapersRecommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabsarXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)