KITE: KV-Invariant Transformer Expansion for Efficient Agentic LLM Scaling
cs.LG, cs.AI
Submitted: 2026-09-23
Updated: 2026-09-23
Terminology
Sources
- GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints
- Program Synthesis with Large Language Models
- Mixture-of-Recursions: Learning Dynamic Recursive Depths for Adaptive Token-Level Computation
- Reducing Transformer Key-Value Cache Size with Cross-Layer Attention
- Evaluating Large Language Models Trained on Code
- Net2Net: Accelerating Learning via Knowledge Transfer
- Training Verifiers to Solve Math Word Problems
- Upcycling Large Language Models into Mixture of Experts
- Measuring Massive Multitask Language Understanding
- Measuring Mathematical Problem Solving With the MATH Dataset
- Training Compute-Optimal Large Language Models
- Scaling Laws for Neural Language Models
- Adam: A Method for Stochastic Optimization
- Sparse Upcycling: Training Mixture-of-Experts from Dense Checkpoints
- Muon is Scalable for LLM Training
- Can a Suit of Armor Conduct Electricity? A New Dataset for Open Book Question Answering
- Fast Transformer Decoding: One Write-Head is All You Need
- Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
- RoFormer: Enhanced Transformer with Rotary Position Embedding
- You Only Cache Once: Decoder-Decoder Architectures for Language Models
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks