IAAIReLIT
Think Deep, Speak Once: Relit, A Recursive Latent Implicit Transformer Framework
Augments a frozen TinyLlama-1.1B with a small trainable recursive block that iteratively refines a latent thought before answering, aiming to combine TRM-style recursive reasoning with the semantics of a pretrained language model.
Inside the method
Frozen LLM → [ Trainable recursive block on z ] × r → answer
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Core recurrence
- Depth control
- Fixed gradient-isolated recurrent loops
- KV / state strategy
- See paper
Reading note
Reports strong parameter efficiency on the GLoRE logical reasoning benchmark, matching or beating larger models on ProofWriter and RuleTaker.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{relit2026,
title = {Think Deep, Speak Once: Relit, A Recursive Latent Implicit Transformer Framework},
author = {Abhishek Panwar and Maheep Singh and Saksham Bansal},
year = {2026},
eprint = {2608.08113},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.08113}
}