NeurIPS 2026 reviews are up: reviews, scores, rebuttals and meta-reviews for 7,877 papers. See the highest rated
acceptodds: the papers of ICLR 2027
470 results for “keyword:"kv"” in ICLR 2027. 87 withdrawn or desk rejected listed apart. clearsort:relevanceratingnewestmost read
[PDF]The Hit Is (Almost) Always Cheaper: Prefix-Cache Economics on Bandwidth-Rich, FLOPs-Poor Accelerators▼1.2k1.67
[PDF]Never Evict Always Taper: A Static Logarithmic Bit-Width Schedule for KV Cache Quantization▲11.00
[PDF]The Sink Tax: Auditing Memory Tokens and Assigning Their Roles in Long-Context Language Models▼4.3k2.00
[PDF]A Memory You Can’t Recall Is No Memory at All: MEMENTO for Query-Agnostic KV Cache Eviction in Multi-Turn Inference▼5.8k2.33
[PDF]Shift-Accumulate Attention: Multiplier-Free Query-Key Products for Transformer Decoding▼4.3k2.00
[PDF]Compression-Aware Fine-Tuning: Adapting Language Model Weights to a Frozen Cache Bottleneck▼4.3k2.00
[PDF]Quantize the Keys, Audit the Values: Retained-State Certificates for KV-Cache Compression▲3952.67
[PDF]RECAP-KV: Single-Prefill Approximation of Reconstruction Importance for Reusable KV Cache Eviction▲5.4k3.00
[PDF]SpecKV: Lossless Speculative Decoding via Dynamic KV Selection for Efficient Inference of Large Vision-Language Models
[PDF]Fast and Quality-Preserving Non-prefix KV Cache Reuse for Large Language Model Serving▼5.8k2.33
[PDF]LPA-CACHE: Latency-Plateau Aware KV Cache Reuse for Efficient Vision-Language-Action Model Inference on GPUs▼1.2k1.67
[PDF]Cache Homeostasis: Diagnosing and Controlling Feature Drift in Autoregressive Video Diffusion▼4.3k2.00
[PDF]REPRISE: Efficient Long-Context Diffusion Inference via Distillation through Reused KV Caches▲5.4k3.00
[PDF]CandiKV: Preserving Candidate KV State Across Selection and Continuation in LLM Workflows▲3952.67
[PDF]Fund Regions, Not Tokens: Query-Conditioned Budget Allocation for Extreme KV-Cache Eviction in Vision-Language Models▲5103.00
[PDF]AVSG: Accelerated Vectorized Sparse Gather for Efficient KV Cache Offload in Sparse-Attention LLM Serving▲9.2k2.67
[PDF]A Memory That Only Looks Like One: A Hybrid Vision-Language Model Relays the Picture Rather Than Storing It▼1.2k1.67
Withdrawn or desk rejected (87)
[PDF]Certified Best-First Sparse Attention: Provable Early Termination via Admissible Block BoundsWithdrawn2.00
[PDF]Cross-Model KV Cache Construction in LLM Familiies: Residual-Stream Mapping Without Matched Head GeometryDesk rejected
[PDF]Trace Forcing: Anchor Tokens in the KV Cache for Training-Free Long Video GenerationWithdrawn2.00
[PDF]EviStream: Evidence-Aligned Audio-Visual Decisions over Partially Observable Video StreamsWithdrawn2.00