Stage-Replay Divergence Follows the KV Cache: Fixed-Prefix Precision Controls and Bidirectional Cache Transplantation
Alexander Boesgaard Lorup
cs.LG, cs.CL
Submitted: 2026-07-30
Comments: 15 pages, 1 figure, 6 tables. Reproducibility artifacts (frozen manifests, token IDs, per-item scores, analysis harnesses) described in Section 3.9
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- The Illusion of Equivalence: Systematic FP16 Divergence in KV-Cached Autoregressive Inference
- MarginGate: Sparse Margin-Triggered Verification for Batch-Invariant LLM Inference
- Beyond Consensus: Trace-Level Synthesis in Mixture of Agents
- SPIRAL: Learning to Search and Aggregate
- Amortized Reasoning Tree Search: Decoupling Proposal and Decision in Large Language Models
- PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning
- Counterfactual Likelihood Tests for Indirect Influence in Private Reasoning Channels
- Re-feeding Is Not Replaying: Measuring Replay Noise in Counterfactual Token-Credit Estimation
- Learning to Reason Across Parallel Samples for LLM Reasoning
- Recursive Self-Aggregation Unlocks Deep Thinking in Large Language Models
- Qwen2.5 Technical Report
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: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks