Tag
reasoning
This tag covers how models reason at inference time, from self re-ranking and shortest-path tasks to recursive reasoning and expert routing in multimodal MoE systems. It matters because small changes in problem length, modality, or routing can expose where reasoning breaks down.
8 articles
Randomized YaRN boosts long-context reasoning
Randomized YaRN helps LLMs generalize better from short training contexts to much longer reasoning windows.
Retrieval that teaches models to reason by analogy
RA-RFT trains retrievers to find useful reasoning analogies, then fine-tunes models with those demonstrations.
Equilibrium Reasoners make latent reasoning scalable
EqR models learn attractors in latent space so iterative reasoning can scale without external verifiers.
AutoTTS lets LLMs discover test-time scaling
AutoTTS turns test-time scaling into an environment search problem, letting LLMs discover cheaper reasoning strategies automatically.
When LLMs Stop Following Procedural Steps
A diagnostic benchmark shows LLMs lose procedural fidelity as step counts grow, even when the arithmetic stays simple.
Select-to-Think: Let SLMs Re-rank Themselves
A new method lets small language models re-rank their own candidates instead of calling an LLM at inference time.
Why LLMs Generalize on Maps but Fail on Scale
A synthetic shortest-path setup shows LLMs transfer across maps, but break when problems get longer because recursive reasoning gets unstable.
Why multimodal MoE models get distracted
A study of multimodal MoE models finds visual inputs can derail routing to reasoning experts, and a routing-guided fix improves results.