Nectar: Neural Estimation of Cached-Token Attention via Regression

• URL: http://arxiv.org/abs/2605.09778v1
• Date: Sun, 10 May 2026 21:51:36 GMT
• Title: Nectar: Neural Estimation of Cached-Token Attention via Regression
• Abstract summary: We show how a compact neural network can be used to evaluate softmax attention over a fixed long context.<n>Nectar fits two networks per layer and KV-head: a target network that predicts the attention output and a score network that predicts the log-normalizer.<n>We report experiments on models from 1.7B to 8B parameters across five long-context datasets.
• Score: 36.709992864731866
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Evaluating softmax attention over a fixed long context requires reading every cached key-value pair for each new query token. For a given context (a book, a manual, a legal corpus) the attention output is a deterministic function of the query. We propose Nectar, which fits a compact neural network to this function for queries drawn from a task-relevant distribution. Nectar fits two networks per layer and KV-head: a target network that predicts the attention output and a score network that predicts the log-normalizer. The pair plugs into the standard masked self-attention at inference time, replacing the $O(n)$ attention over the cache with a forward pass whose cost does not depend on $n$. Each module carries on the order of $|θ|$ parameters per layer and KV-head, typically much smaller than the $2nd$ KV-cache footprint at the same granularity. We report experiments on models from 1.7B to 8B parameters across five long-context datasets. The approximation error tracks the next-token accuracy gap to full attention, and allocating capacity non-uniformly across layers reduces that gap in our ablation. Beyond this analysis of metrics, we check that the text generations (following a question prompt) of a model equipped with a Nectar module match in semantic content those obtained by giving the same model access to the full cache.

Related papers

• SSTQ:Privacy-Preserving Vector Quantization via Subsampled Stochastic TurboQuant [79.24089819400126]
  Subsampled TurboQuant (SSTQ) is a framework that combines overcomplete equal-width tight frames, coordinate subsampling, and privacy-aware quantization.<n>SSTQ achieves mean squared error scaling while using only $lceil log N il + b$ bits per client, where $N = (d)$ is the frame size.<n>We also derive a surrogate privacy-aware codebook objective that reduces the codebook-dependent MSE scaling from $O(4b)$ to $O(2b)$.
  arXiv  Detail & Related papers  (2026-08-05T17:51:25Z)
• Back from the Future: Key-Value Cache Management by Counter-Causal Surprise [67.1056509495879]
  Key-value (KV) cache management through compression and eviction strategies has emerged as an important research direction in recent years.<n>We propose a simple yet effective KV eviction scheme motivated by the insight that past tokens which can be well-predicted from more recent tokens are redundant.<n>We evaluate our strategy on various open-source LLMs and benchmark datasets showing competitive or improved performance over other state-of-the-art methods.
  arXiv  Detail & Related papers  (2026-07-30T02:42:51Z)
• Saturation Makes Quantization Error Additive: A Coverage Model with a Certificate [0.0]
  Mixed-precision quantization must decide which parts of a model to keep at higher precision.<n>We show that the loss from quantizing a set of layers can be reconstructed from per-layer or pairwise sensitivities measured in isolation.<n>We propose the coverage model $f(S)=cbigl (1-prod_iin S (1-a_i)bigr)$, which reproduces the measured variance profile of $f$ to within a few percent from its $L$ fitted break-rates.
  arXiv  Detail & Related papers  (2026-07-14T02:08:33Z)
• Stochastic Sparse Attention for Memory-Bound Inference [19.301894658575502]
  We present Additive No-mulT Attention (SANTA), a method that sparsifies value-cache access by sampling $S ll n_k$ indices from the post-softmax distribution.<n>We also propose Bernoulli $qKmathsfT$ sampling as a complementary technique to sparsify the score stage.
  arXiv  Detail & Related papers  (2026-05-03T14:44:14Z)
• EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models [8.323540970510809]
  We propose EntropyCache, a training-free KV caching method that uses the maximum entropy of newly decoded token distributions as a constant-cost signal for deciding when to recompute.<n> Experiments on LLaDA-8B-Instruct and Dream-7B-Instruct show that EntropyCache achieves $15.2times$-$26.4times$ speedup on standard benchmarks and $22.4times$-$24.1times$ on chain-of-thought benchmarks, with competitive accuracy and decision overhead accounting for only $0.5%$ of inference time.
  arXiv  Detail & Related papers  (2026-03-19T04:46:34Z)
• Catch Your Breath: Adaptive Computation for Self-Paced Sequence Production [55.76222360698305]
  We explore a class of supervised training objectives that allow a language model to dynamically and autonomously scale the number of compute steps used for each input token.<n>For any token, the model can request additional compute steps by emitting a don't know> output.<n>We find that the CYB model requests additional steps when doing so improves accuracy, and the model adapts its processing time to token-level complexity and context.
  arXiv  Detail & Related papers  (2025-10-13T21:07:05Z)
• DISCO: Diversifying Sample Condensation for Efficient Model Evaluation [59.01400190971061]
  Costly evaluation reduces inclusivity, slows the cycle of innovation, and worsens environmental impact.<n>We argue that promoting diversity among samples is not essential; what matters is to select samples thatmaximise diversity in model responses.<n>Our method, $textbfDiversifying Sample Condensation (DISCO)$, selects the top-k samples with the greatest model disagreements.
  arXiv  Detail & Related papers  (2025-10-09T08:53:59Z)
• Certifiably Robust Model Evaluation in Federated Learning under Meta-Distributional Shifts [8.700087812420687]
  We provide guarantees for the model's performance on a different, unseen network "B"<n>We show how the principled vanilla DKW bound enables certification of the model's true performance on unseen clients within the same (source) network.
  arXiv  Detail & Related papers  (2024-10-26T18:45:15Z)
• Semi-Supervised and Long-Tailed Object Detection with CascadeMatch [91.86787064083012]
  We propose a novel pseudo-labeling-based detector called CascadeMatch. Our detector features a cascade network architecture, which has multi-stage detection heads with progressive confidence thresholds. We show that CascadeMatch surpasses existing state-of-the-art semi-supervised approaches in handling long-tailed object detection.
  arXiv  Detail & Related papers  (2023-05-24T07:09:25Z)
• Fundamental tradeoffs between memorization and robustness in random features and neural tangent regimes [15.76663241036412]
  We prove for a large class of activation functions that, if the model memorizes even a fraction of the training, then its Sobolev-seminorm is lower-bounded. Experiments reveal for the first time, (iv) a multiple-descent phenomenon in the robustness of the min-norm interpolator.
  arXiv  Detail & Related papers  (2021-06-04T17:52:50Z)
• A Law of Robustness for Weight-bounded Neural Networks [37.54604146791085]
  Recently, (Bubeck et al., 2020) conjectured that when using two-layer networks with $k$ neurons to fit a generic dataset, the smallest Lipschitz constant is $Omega(sqrtfracnk)$. In this work we derive a lower bound on the Lipschitz constant for any arbitrary model class with bounded Rademacher complexity. Our result coincides with that conjectured in (Bubeck et al., 2020) for two-layer networks under the assumption of bounded weights.
  arXiv  Detail & Related papers  (2021-02-16T11:28:59Z)
• $O(n)$ Connections are Expressive Enough: Universal Approximability of Sparse Transformers [71.31712741938837]
  We show that sparse Transformers with only $O(n)$ connections per attention layer can approximate the same function class as the dense model with $n2$ connections. We also present experiments comparing different patterns/levels of sparsity on standard NLP tasks.
  arXiv  Detail & Related papers  (2020-06-08T18:30:12Z)
• Cryptanalytic Extraction of Neural Network Models [56.738871473622865]
  We introduce a differential attack that can efficiently steal the parameters of the remote model up to floating point precision. Our attack relies on the fact that ReLU neural networks are piecewise linear functions. We extract models that are 220 times more precise and require 100x fewer queries than prior work.
  arXiv  Detail & Related papers  (2020-03-10T17:57:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.