REVIEW 48 cited by
Hungry Hungry Hippos: Towards Language Modeling with State Space 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
Hungry Hungry Hippos: Towards Language Modeling with State Space Models
read the original abstract
State space models (SSMs) have demonstrated state-of-the-art sequence modeling performance in some modalities, but underperform attention in language modeling. Moreover, despite scaling nearly linearly in sequence length instead of quadratically, SSMs are still slower than Transformers due to poor hardware utilization. In this paper, we make progress on understanding the expressivity gap between SSMs and attention in language modeling, and on reducing the hardware barrier between SSMs and attention. First, we use synthetic language modeling tasks to understand the gap between SSMs and attention. We find that existing SSMs struggle with two capabilities: recalling earlier tokens in the sequence and comparing tokens across the sequence. To understand the impact on language modeling, we propose a new SSM layer, H3, that is explicitly designed for these abilities. H3 matches attention on the synthetic languages and comes within 0.4 PPL of Transformers on OpenWebText. Furthermore, a hybrid 125M-parameter H3-attention model that retains two attention layers surprisingly outperforms Transformers on OpenWebText by 1.0 PPL. Next, to improve the efficiency of training SSMs on modern hardware, we propose FlashConv. FlashConv uses a fused block FFT algorithm to improve efficiency on sequences up to 8K, and introduces a novel state passing algorithm that exploits the recurrent properties of SSMs to scale to longer sequences. FlashConv yields 2$\times$ speedup on the long-range arena benchmark and allows hybrid language models to generate text 2.4$\times$ faster than Transformers. Using FlashConv, we scale hybrid H3-attention language models up to 2.7B parameters on the Pile and find promising initial results, achieving lower perplexity than Transformers and outperforming Transformers in zero- and few-shot learning on a majority of tasks in the SuperGLUE benchmark.
Forward citations
Cited by 48 Pith papers
-
Structure Before Collapse: Transient semantic geometry in next-token prediction
Semantic geometry emerges transiently early in next-token prediction training before collapsing to Neural Collapse symmetry in synthetic settings with latent semantic factors.
-
Frequency Domain Reservoir Computing
FRESCO is a frequency-domain Echo State Network using zero-padding embeddings, packed readout, and native frequency non-linearity to achieve O(N) complexity while matching SOTA on memory and forecasting benchmarks.
-
LongSpike: Fractional Order Spiking State Space Models for Efficient Long Sequence Learning
LongSpike integrates fractional-order state-space modeling into spiking neural networks, enabling better long-sequence performance than prior SNNs on LRA, WikiText-103, and Speech Commands benchmarks while retaining s...
-
TCP-SSM: Efficient Vision State Space Models with Token-Conditioned Poles
TCP-SSM conditions stable poles on visual tokens to explicitly control memory decay and oscillation in SSMs, cutting computation up to 44% while matching or exceeding accuracy on classification, segmentation, and detection.
-
Exact Flow Linear Attention: Exact Solution from Continuous-Time Dynamics
Exact Flow Linear Attention derives a closed-form exact update for delta-rule linear attention from continuous-time dynamics, removing Euler discretization error while preserving linear complexity and structure.
-
Griffin: Mixing Gated Linear Recurrences with Local Attention for Efficient Language Models
Griffin hybrid model matches Llama-2 performance while trained on over 6 times fewer tokens and offers lower inference latency with higher throughput.
-
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...
-
From Tensor Buffer to Distributed Memory Hierarchy: A Survey of KV Cache Management for LLM Serving
KV-cache serving systems concentrate into five archetypes under a four-axis taxonomy, with ownership explaining residual distributed design variance and seven measurement gaps blocking next steps.
-
Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity
SDM sparsifies the Gated DeltaNet update rule to enable 1000x larger recurrent memory states at iso-FLOP, improving long-context recall and short-context reasoning over GDN and matching full attention at 8B scale.
-
UniSkip-Mamba: A Frequency-Aware State Space Model for Audio-Visual Temporal Forgery Localization
A stride-2 skip-scanning Mamba with temporally concatenated audio-visual features sets new state-of-the-art scores on LAV-DF and AV-Deepfake1M temporal forgery localization.
-
UniSkip-Mamba: A Frequency-Aware State Space Model for Audio-Visual Temporal Forgery Localization
Skip-scanning Mamba with unified audio-visual sequences reaches 63.4% AP@0.95 on LAV-DF and 63.58% mAP on AV-Deepfake1M by regularizing toward low/mid-frequency forgery cues.
-
Spatial-Aware Reduction Framework: Towards Efficient and Faithful Visual State Space Models
STORM is a training-free spatial-aware token reduction framework that reformulates compression on spatial units to preserve grid topology and neighborhood coherence in visual state space models.
-
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.
-
Structured Recurrent Mixers for Massively Parallelized Sequence Generation
Structured Recurrent Mixers provide a dual parallel-recurrent representation for sequence models, claiming superior training efficiency, information capacity, and inference throughput over linear complexity alternatives.
-
Structured Recurrent Mixers for Massively Parallelized Sequence Generation
Structured Recurrent Mixers enable algebraic switching between parallel training and recurrent inference representations, yielding higher throughput, concurrency, and training efficiency than comparable linear-complex...
-
Structured Recurrent Mixers for Massively Parallelized Sequence Generation
Structured Recurrent Mixers enable algebraic switching between parallel training and recurrent inference representations, delivering higher efficiency, information capacity, and throughput than other linear-complexity models.
-
ZAYA1-8B Technical Report
ZAYA1-8B is a reasoning MoE model with 700M active parameters that matches larger models on math and coding benchmarks and reaches 91.9% on AIME'25 via Markovian RSA test-time compute.
-
The Impossibility Triangle of Long-Context Modeling
No model can achieve efficiency, compactness, and recall capacity scaling with sequence length at once, as any two imply a strict bound of O(poly(d)/log V) on recallable facts.
-
CLIMB: Controllable Longitudinal Brain Image Generation using Mamba-based Latent Diffusion Model and Gaussian-aligned Autoencoder
CLIMB generates controllable longitudinal brain MRI images from baseline scans using a Mamba-based latent diffusion model and Gaussian-aligned autoencoder, reporting SSIM 0.9433 on the ADNI dataset of 6306 scans.
-
M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling
M²RNN achieves perfect state tracking at unseen lengths and outperforms Gated DeltaNet hybrids by 0.4-0.5 perplexity on 7B models with 3x smaller recurrent states.
-
Rethinking Efficiency in Neural Combinatorial Optimization: Batched Preference Optimization with Mamba
ECO combines a Mamba encoder-decoder with two-stage batched DPO and LS-aware preference construction, claiming the best reported neural performance and near-linear memory scaling on TSP up to 5000 nodes and CVRP up to...
-
Rethinking Efficiency in Neural Combinatorial Optimization: Batched Preference Optimization with Mamba
ECO uses supervised warm-up plus iterative batched DPO on a Mamba backbone to reach top neural performance on TSP and CVRP while lowering memory growth and raising throughput.
-
Partial Ring Scan: Revisiting Scan Order in Vision State Space Models
Ring-based scanning with selective channel routing improves accuracy, speed, and rotation robustness of vision state-space models.
-
Physics of Language Models: Part 4.1, Architecture Design and the Magic of Canon Layers
Canon layers—residual 1-d causal convolutions over adjacent tokens—boost synthetic reasoning depth 2-4x, lift NoPE to RoPE level, and bring GLA up to Mamba2/GDN, with qualitative real-world confirmation.
-
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.
-
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.
-
Short window attention enables long-term memorization
Short sliding windows in hybrid attention-xLSTM models boost long-context performance by encouraging long-term memory use, and stochastic window sizing improves both short and long tasks.
-
RoboSSM: Scalable In-context Imitation Learning via State-Space Models
RoboSSM shows that a state-space model backbone can extend in-context imitation learning to prompts much longer than those seen in training, where a Transformer-based baseline degrades.
-
Customizing the Inductive Biases of Softmax Attention using Structured Matrices
Structured-matrix scoring functions, BTT and MLR, let attention escape the low-rank bottleneck and add a distance-dependent compute bias, improving accuracy for fixed compute on regression, language modeling, and forecasting.
-
mGRADE: Minimal Recurrent Gating Meets Delay Convolutions for Lightweight Sequence Modeling
mGRADE uses learnable-spaced convolutions shown to be equivalent to delay embeddings plus a lightweight gated recurrent component to achieve low-memory multi-timescale sequence modeling.
-
CodeBrain: Bridging Decoupled Tokenizer and Multi-Scale Architecture for EEG Foundation Model
CodeBrain introduces a decoupled TFDual-Tokenizer and multi-scale EEGSSM architecture for an EEG foundation model pretrained on a large corpus, claiming strong generalization across eight downstream tasks and ten datasets.
-
Retentive Network: A Successor to Transformer for Large Language Models
RetNet is a new sequence modeling architecture that delivers parallel training, constant-time inference, and competitive language modeling performance as a potential replacement for Transformers.
-
ZUNA1.1: A more flexible EEG foundation model for Denoising and Super-resolution
ZUNA1.1, an open-source 380M EEG diffusion autoencoder, reconstructs variable-length, flexibly masked EEG at least as well as its predecessor and far better than spherical spline interpolation.
-
SketchMamba: A Lightweight State-Space Model for Joint Progressive Sketch Classification and Stroke Auto-Completion
Dense per-step class loss on a 1.55M-parameter causal Mamba unifies progressive sketch recognition (94.93% final Top-1, 0.706 progressive AUC) and stroke auto-completion on 58-class Quick, Draw!.
-
CogSENet: Blind Image Deblurring with Blur-Conditioned Semantic Routing and Explicit Frequency Fusion
CogSENet proposes semantic-driven state space modules, bi-frequency fusion blocks, and continuous blur field estimation to outperform prior blind deblurring methods with fewer parameters.
-
How Optimality Structures Sparse Dictionaries: A Theory for Understanding SAE Representations
Derives optimality constraints for nonnegative joint dictionary learning that explain observed SAE behaviors such as feature splitting, absorption, and dense antipodal features.
-
Latent Recurrent Transformer: Architecture Exploration, Training Strategies, and Scaling Behavior
Latent Recurrent Transformer augments autoregressive transformers with a cross-layer recurrent latent pathway from prior hidden states and uses interleaved parallel training to improve loss and in-context learning at ...
-
Structured Recurrent Mixers for Massively Parallelized Sequence Generation
By restricting token mixing to rank-1 decayed matrices, the same model trains in parallel but generates recurrently with constant memory, yielding large throughput gains in compute-matched benchmarks.
-
Kaczmarz Linear Attention
Kaczmarz Linear Attention replaces the empirical coefficient in Gated DeltaNet with a key-norm-normalized step size derived from the online regression objective, yielding lower perplexity and better needle-in-haystack...
-
mHC-SSM: Manifold-Constrained Hyper-Connections for State Space Language Models with Stream-Specialized Adapters
Manifold-constrained multi-stream mixing plus per-stream adapters improves SSM language model validation loss from 6.3507 to 6.1353 and perplexity from 572.91 to 461.88 on WikiText-2.
-
Toeplitz MLP Mixers are Low Complexity, Information-Rich Sequence Models
Toeplitz MLP Mixers replace attention with masked Toeplitz multiplications for sub-quadratic complexity while retaining more sequence information and outperforming on copying and in-context tasks.
-
Distill-then-Replace: Efficient Task-Specific Hybrid Attention Model Construction
Distill-then-Replace builds task-specific hybrid attention LLMs by distilling each full-attention block into a linear counterpart and greedily replacing layers under a validation-performance constraint.
-
EventCrab: Harnessing Frame and Point Synergy for Event-based Action Recognition and Beyond
EventCrab integrates frame and point networks with a joint representation space, SCL, and Hilbert-scan EPE to improve event-based action recognition by 5-7% on two datasets.
-
ZONOS2 Technical Report
ZONOS2 8B is a scaled MoE TTS model with 900M active parameters trained on 6M hours of data that reports competitive SOTA results on naturalness, speaker similarity, WER, and a new ZTTS1-Eval benchmark while releasing...
-
Hybrid Architectures for Language Models: Systematic Analysis and Design Insights
This work systematically compares inter-layer and intra-layer hybridization strategies for combining self-attention and Mamba-style state space models, evaluating them on language modeling, downstream tasks, long-cont...
-
ZONOS2 Technical Report
ZONOS2 8B scales a prior TTS system to 8B parameters with MoE architecture and 6M hours of data, reporting competitive benchmark performance on naturalness and speaker similarity while releasing weights.
-
A Survey on Efficient Inference for Large Language Models
The paper surveys techniques to speed up and reduce the resource needs of LLM inference, organized by data-level, model-level, and system-level changes, with comparative experiments on representative methods.
-
Advancing Intelligent Sequence Modeling: Evolution, Trade-offs, and Applications of State- Space Architectures from S4 to Mamba
A survey tracing the evolution of state-space models like S4 and Mamba, their efficiency trade-offs, and applications in NLP, vision, and other domains.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.