MultiHashFormer enables hash-based autoregression in LMs by encoding tokens as multi-hash signatures, outperforming standard Transformers at 100M-3B scales while keeping parameter count constant for multilingual expansion.
hub
Liu, and Matt Gardner
22 Pith papers cite this work. Polarity classification is still indexing.
22
Pith papers citing it
hub tools
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
AuthorityBench shows citation presence (real or fabricated) increases LLM hallucination rates vs no-citation baseline, strongest for fabricated citations on true claims, with domain variation but negligible venue or author effects.
SemGrad measures LLM uncertainty via gradients in semantic space using a Semantic Preservation Score to select embeddings, with HybridGrad combining it with parameter gradients to outperform sampling-based baselines especially when multiple responses are valid.
Preconditioned delta-rule models with a diagonal curvature approximation improve upon standard DeltaNet, GDN, and KDA by better approximating the test-time regression objective.
Answer tokens show forward drift and key-anchor focus when reading correct reasoning traces; a geometric-plus-semantic SRQ steering method boosts quantitative reasoning accuracy without training.
RWKV uses a linear attention mechanism to deliver Transformer-level performance with RNN-style inference efficiency, demonstrated at up to 14 billion parameters.
A two-stream Transformer variant that separates state storage from next-token prediction improves validation loss and downstream task performance by 2-3 points over standard Transformers.
Manifold Power Iteration aligns MoE router rows with principal singular directions of experts via a power-then-retract process, with theory showing convergence and experiments on 1B-11B models showing gains.
LINK improves cross-lingual knowledge transfer via lexical substitutions in English pretraining data, yielding notable downstream gains and up to 2x training speedup across eight languages and five model sizes.
Empirical study shows mixture pretraining tolerates higher target data repetition than single-source training, with a new repetition-aware scaling law enabling principled mixture selection based on data size, compute, and model scale.
A shared global expert pool in MoE improves validation loss over per-layer experts and allows sublinear expert-parameter growth with depth.
GRASPrune removes 50% of parameters from LLaMA-2-7B via global gating and projected straight-through estimation, reaching 12.18 WikiText-2 perplexity and competitive zero-shot accuracy after four epochs on 512 calibration sequences.
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
Adapts multi-layer token-level Mahalanobis distance with supervised linear regression to yield improved uncertainty scores for LLM truthfulness tasks.
A regression model using attention features and recurrent uncertainty scores improves selective generation in LLMs over unsupervised and supervised baselines on ten datasets and three models.
GPT-NeoX-20B is a publicly released 20B parameter autoregressive language model trained on the Pile that shows strong gains in five-shot reasoning over similarly sized prior models.
A factorized study finds raw hidden states and attention features hard to beat in-domain for LLM uncertainty probes, but structured compressed features are more robust under distribution shift, with pretrained probes transferring to open-ended generation.
GRASP is a scalable method for subset-level data attribution in pretraining that models interactions via a geometry-aware quadratic penalty and claims to double rank correlation while cutting costs.
q0 turns multi-epoch budgets into diverse model populations using three primitives that outperform single-model training and strong ensembles with fewer epochs on a 1.8B model.
SIVR detects LLM hallucinations by learning from token-wise and layer-wise variance patterns in internal hidden states, outperforming baselines with better generalization and less training data.
Learning rate decay during SFT increases pretrained model sharpness, which exacerbates catastrophic forgetting and causes overtraining in LLMs.
LLM accuracy on reasoning tasks differs significantly by question type, with step-by-step reasoning accuracy often uncorrelated to final answer selection.
citing papers explorer
-
MultiHashFormer: Hash-based Generative Language Models
MultiHashFormer enables hash-based autoregression in LMs by encoding tokens as multi-hash signatures, outperforming standard Transformers at 100M-3B scales while keeping parameter count constant for multilingual expansion.
-
Authority, Truth, and Citation Bias: A Large-Scale Multi-Domain Benchmark for Studying Epistemic Susceptibility in Large Language Models
AuthorityBench shows citation presence (real or fabricated) increases LLM hallucination rates vs no-citation baseline, strongest for fabricated citations on true claims, with domain variation but negligible venue or author effects.
-
Gradients with Respect to Semantics Preserving Embeddings Tell the Uncertainty of Large Language Models
SemGrad measures LLM uncertainty via gradients in semantic space using a Semantic Preservation Score to select embeddings, with HybridGrad combining it with parameter gradients to outperform sampling-based baselines especially when multiple responses are valid.
-
Preconditioned DeltaNet: Curvature-aware Sequence Modeling for Linear Recurrences
Preconditioned delta-rule models with a diagonal curvature approximation improve upon standard DeltaNet, GDN, and KDA by better approximating the test-time regression objective.
-
How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning
Answer tokens show forward drift and key-anchor focus when reading correct reasoning traces; a geometric-plus-semantic SRQ steering method boosts quantitative reasoning accuracy without training.
-
RWKV: Reinventing RNNs for the Transformer Era
RWKV uses a linear attention mechanism to deliver Transformer-level performance with RNN-style inference efficiency, demonstrated at up to 14 billion parameters.
-
The State-Prediction Separation Hypothesis
A two-stream Transformer variant that separates state storage from next-token prediction improves validation loss and downstream task performance by 2-3 points over standard Transformers.
-
Redesign Mixture-of-Experts Routers with Manifold Power Iteration
Manifold Power Iteration aligns MoE router rows with principal singular directions of experts via a power-then-retract process, with theory showing convergence and experiments on 1B-11B models showing gains.
-
Multilingual Knowledge Transfer under Data Constraints via Lexical Interventions
LINK improves cross-lingual knowledge transfer via lexical substitutions in English pretraining data, yielding notable downstream gains and up to 2x training speedup across eight languages and five model sizes.
-
Scaling Laws for Mixture Pretraining Under Data Constraints
Empirical study shows mixture pretraining tolerates higher target data repetition than single-source training, with a new repetition-aware scaling law enabling principled mixture selection based on data size, compute, and model scale.
-
UniPool: A Globally Shared Expert Pool for Mixture-of-Experts
A shared global expert pool in MoE improves validation loss over per-layer experts and allows sublinear expert-parameter growth with depth.
-
GRASPrune: Global Gating for Budgeted Structured Pruning of Large Language Models
GRASPrune removes 50% of parameters from LLaMA-2-7B via global gating and projected straight-through estimation, reaching 12.18 WikiText-2 perplexity and competitive zero-shot accuracy after four epochs on 512 calibration sequences.
-
Token-Level Density-Based Uncertainty Quantification Methods for Eliciting Truthfulness of Large Language Models
Adapts multi-layer token-level Mahalanobis distance with supervised linear regression to yield improved uncertainty scores for LLM truthfulness tasks.
-
Unconditional Truthfulness: Learning Unconditional Uncertainty of Large Language Models
A regression model using attention features and recurrent uncertainty scores improves selective generation in LLMs over unsupervised and supervised baselines on ten datasets and three models.
-
GRASP: Geometry-aware Residual Alignment for Scalable Pretraining Data Attribution
GRASP is a scalable method for subset-level data attribution in pretraining that models interactions via a geometry-aware quadratic penalty and claims to double rank correlation while cutting costs.
-
q0: Primitives for Hyper-Epoch Pretraining
q0 turns multi-epoch budgets into diverse model populations using three primitives that outperform single-model training and strong ensembles with fewer epochs on a 1.8B model.
-
Learning Uncertainty from Sequential Internal Dispersion in Large Language Models
SIVR detects LLM hallucinations by learning from token-wise and layer-wise variance patterns in internal hidden states, outperforming baselines with better generalization and less training data.
-
(How) Learning Rates Regulate Catastrophic Overtraining
Learning rate decay during SFT increases pretrained model sharpness, which exacerbates catastrophic forgetting and causes overtraining in LLMs.
-
Is Large Language Model Performance on Reasoning Tasks Impacted by Different Ways Questions Are Asked?
LLM accuracy on reasoning tasks differs significantly by question type, with step-by-step reasoning accuracy often uncorrelated to final answer selection.