Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models
Paper • 2605.07721 • Published • 29
Thomas
tomsal
AI & ML interests
Reasoning, LLM, Generative Models, AI4Science
Recent Activity
upvoted a paper about 2 months ago
Reasoning as Compression: Unifying Budget Forcing via the Conditional Information Bottleneck upvoted a paper 2 months ago
Efficient Training-Free Multi-Token Prediction via Embedding-Space ProbingOrganizations
None yet