REVIEW 31 cited by
Zoology: Measuring and Improving Recall in Efficient Language Models
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Zoology: Measuring and Improving Recall in Efficient Language Models
read the original abstract
Attention-free language models that combine gating and convolutions are growing in popularity due to their efficiency and increasingly competitive performance. To better understand these architectures, we pretrain a suite of 17 attention and "gated-convolution" language models, finding that SoTA gated-convolution architectures still underperform attention by up to 2.1 perplexity points on the Pile. In fine-grained analysis, we find 82% of the gap is explained by each model's ability to recall information that is previously mentioned in-context, e.g. "Hakuna Matata means no worries Hakuna Matata it means no" $\rightarrow$ "??". On this task, termed "associative recall", we find that attention outperforms gated-convolutions by a large margin: a 70M parameter attention model outperforms a 1.4 billion parameter gated-convolution model on associative recall. This is surprising because prior work shows gated convolutions can perfectly solve synthetic tests for AR capability. To close the gap between synthetics and real language, we develop a new formalization of the task called multi-query associative recall (MQAR) that better reflects actual language. We perform an empirical and theoretical study of MQAR that elucidates differences in the parameter-efficiency of attention and gated-convolution recall. Informed by our analysis, we evaluate simple convolution-attention hybrids and show that hybrids with input-dependent sparse attention patterns can close 97.4% of the gap to attention, while maintaining sub-quadratic scaling. Our code is accessible at: https://github.com/HazyResearch/zoology.
Forward citations
Cited by 31 Pith papers
-
Learning What Matters: Supervising Global Context Pruning with Causal Evidence Sets
Attention is an unreliable supervision target for context pruning; causal evidence sets recovered by masking produce routers that match or beat attention-distilled routers, especially on multi-hop and conflicting-fact tasks.
-
The Orthogonalized Read Is a Removable Training Scaffold for Recurrent Memory
The orthogonalized read on noisy recall is a removable training scaffold: it multiplies escape hazard ~6.4x during plateau training, but a stock mLSTM achieves full accuracy once the read is annealed away.
-
Parallax: Parameterized Local Linear Attention for Language Modeling
Parallax is a scalable parameterized local linear attention variant that improves LLM pretraining perplexity at 0.6B/1.7B scales with a hardware-aware kernel and shows gains under parameter- and compute-matched controls.
-
Variational Linear Attention: Stable Associative Memory for Long-Context Transformers
VLA stabilizes linear attention by solving regularized least-squares updates with unit-length writes, yielding Jacobian spectral norm exactly 1 and 109x smaller state norms while improving multi-query recall accuracy ...
-
Long Context Pre-Training with Lighthouse Attention
Lighthouse Attention enables faster long-context pre-training via gradient-free symmetrical hierarchical compression of QKV while preserving causality, followed by a short full-attention recovery that yields lower los...
-
DSSMs: State Space Models with Explicit Memory via Delay Differential Equations
Delay State Space Models augment diagonal SSMs with explicit delayed feedback, stable discrete parameterization, and FFT training, improving delayed-retrieval tasks and matching or beating S4D on most standard sequenc...
-
The Capability Convergence Hypothesis: Capability from Access Structure, Not Scale
Long-context capability converges to hybrid architectures pairing O(1) state with a scalable index; pure scaling of SSMs or Transformers alone hits information, horizon, or circuit walls.
-
The Capability Convergence Hypothesis: Capability from Access Structure, Not Scale
Capability, unlike representation, converges to access-complete hybrid architectures, and a witness task separates pure from hybrid families under fixed budgets.
-
Remembering Distinct Items, Not Tokens: A Learnable Dirichlet-Process Cache Between State-Space Models and Attention
A DP-means allocate-on-novelty cache matches full-attention associative recall while storing only distinct items, and a minimal novelty gate recovers the rule end-to-end.
-
A Hippocampus for Linear Attention: An Exact Memory for What the Recurrent State Forgets
HOLA pairs a compressive delta-rule recurrent state with a residual-selected exact KV cache and decoupled RMSNorm-gamma read, yielding lower perplexity than both standard linear attention and full-attention baselines ...
-
Dynamic Short Convolutions Improve Transformers
Dynamic short convolutions applied to key/query/value projections and linear layers in Transformers yield consistent performance gains and 1.33-1.60x compute advantages over standard models on language modeling from 1...
-
Blurry Window Attention
Blurry Window Attention stores a frequency window and reconstructs blurry KV history via Dirichlet kernel interpolation, achieving 8x better state efficiency than sliding window attention on the MQAR synthetic task.
-
Memory by Design: Probabilistic Sequence Layers
The design-model framework unifies sub-quadratic sequence models as Bayesian filters and introduces a covariance-tracking Bayesian Layer that improves retrieval robustness beyond training regimes on MQAR and RULER benchmarks.
-
OSDN: Improving Delta Rule with Provable Online Preconditioning in Linear Attention
OSDN adds online diagonal preconditioning to the Delta Rule, preserving chunkwise parallelism while proving super-geometric convergence and delivering 32-39% recall gains at 340M-1.3B scales.
-
HubRouter: A Pluggable Sub-Quadratic Routing Primitive for Hybrid Sequence Models
HubRouter is a sub-quadratic routing primitive using learned hubs that replaces attention layers in hybrid models while delivering competitive perplexity and large throughput gains.
-
Kalman Linear Attention: Parallel Bayesian Filtering For Efficient Language Modelling and State Tracking
KLA reparameterises information-form Kalman filtering as an associative Möbius scan, giving a parallelisable nonlinear state-space mixer that tracks A5 permutations with one or two layers.
-
Towards Understanding What State Space Models Learn About Code
SSM code models capture code syntax and semantics better than Transformers before fine-tuning, forget short-range structure when fine-tuned on type inference, and an added high-frequency path or more kernels recovers ...
-
Gated KalmaNet: A Fading Memory Layer Through Test-Time Ridge Regression
Gated KalmaNet uses exact Kalman gain computation with adaptive gating and Chebyshev iteration to improve SSM performance on long-context tasks over prior approximations like DeltaNet.
-
Controllably Efficient Language Models
A single transformer variant can compress past context into chunk summaries and use chunk size as a test-time knob to trade quality against speed and memory, outperforming many efficient baselines on recall benchmarks.
-
Kimi Linear: An Expressive, Efficient Attention Architecture
Kimi Linear hybridizes linear attention with a new KDA module to beat full attention on tasks while slashing KV cache by 75% and speeding decoding up to 6x.
-
Eigenvalues as a Metric for Memory Dynamics in Sequence Models
Eigenvalue spectra of attention and SSM dynamics show consistent signatures of memory retention and selective forgetting that align with task requirements.
-
An Empirical Study of Mamba-based Language Models
An 8B Mamba-2-Hybrid with 43% Mamba-2, 7% attention, and 50% MLP layers exceeds an 8B Transformer by 2.65 points on average across 12 tasks and matches it on 23 long-context tasks while enabling up to 8x faster inference.
-
Gated Linear Attention Transformers with Hardware-Efficient Training
Gated linear attention Transformers achieve competitive language modeling results with linear-time inference, superior length generalization, and higher training throughput than Mamba.
-
Context by Distinct Information: An Auditable Dirichlet-Process Working Memory for Long, Redundant Context Streams
A Dirichlet-process novelty cache plus recurrence matches full-attention coupling at about half the attended tokens on long redundant streams, with inspectable retained slots.
-
Q-Delta: Beyond Key-Value Associative State Evolution
Q-Delta extends linear attention by introducing a query-conditioned delta rule that incorporates mixed key-query errors into recurrent state updates for improved stability and performance.
-
Adaptive Memory Decay for Log-Linear Attention
Making memory decay input-dependent via a lightweight MLP improves log-linear attention performance on associative recall, selective copying, and language modeling, especially for long sequences.
-
MDN: Parallelizing Stepwise Momentum for Delta Linear Attention
MDN parallelizes stepwise momentum for delta linear attention using geometric reordering and dynamical systems analysis, yielding performance gains over Mamba2 and GDN on 400M and 1.3B models.
-
Sessa: Selective State Space Attention
Sessa integrates attention within recurrent paths to achieve power-law memory tails and flexible non-decaying selective retrieval, outperforming baselines on long-context tasks.
-
Gated Delta Networks: Improving Mamba2 with Delta Rule
Gated DeltaNet integrates gating and delta rules into linear transformers, outperforming Mamba2 and DeltaNet on language modeling, reasoning, retrieval, and long-context tasks.
-
Naju: A Native Discrete State-Space Model with Independent Retention and Writing for Long-Sequence Memory
Decoupling the forget gate from the input gate in a discrete state-space recurrence lets one fixed-size state both preserve old bindings over long horizons and overwrite stale ones.
-
Convolution for Large Language Models
Adding a residual depthwise convolution (kernel 3) to QKV projections before attention raises average downstream accuracy in Qwen3-1.7B/4B by 1.6-3.8 points with negligible parameter cost.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.