FASER delivers up to 53% higher throughput and 1.92x lower latency in dynamic LLM serving by adjusting speculative lengths per request, early pruning of rejects, and overlapping draft/verification phases via frontiers.
arXiv preprint arXiv:2403.02181 , year=
10 Pith papers cite this work. Polarity classification is still indexing.
10
Pith papers citing it
citation-role summary
method 1
citation-polarity summary
verdicts
UNVERDICTED 10roles
method 1polarities
use method 1representative citing papers
Coordinating layer-wise and sentence-wise early exits in LLMs produces multiplicative speedups of 1.4-2.3x over single-dimension early exit on sentiment classification tasks.
L2A trains one LLM with input-and-budget-conditioned gates to adapt sparsity across layers, heads, and tokens, tracing the compute-accuracy frontier while staying within 0.6% of dense performance at 34% layer sparsity on tested models.
LLM computation follows a consistent two-phase pattern: a sparse early-layer core reconstructs the head of the output distribution, with later layers and attention heads providing incremental refinements that correlate with model uncertainty.
Introduces LOES, a constructive spectral method to select task-discriminative subspaces from intermediate layer embeddings, and GeoReg for enforcing simplicial class geometry during fine-tuning, with reported gains increasing with model depth across modalities.
InfoRidge reveals a non-monotonic pattern in which predictive mutual information between hidden states and outputs peaks in intermediate layers before declining in final layers.
EASE-TTT creates a soft attention target from evidence chunks to guide query-side test-time adaptation, yielding higher macro-average scores than full-context, retrieval-only, and standard qTTT baselines on six LongBench QA tasks.
BMCR uses RL to adaptively compose modules from CNN and ViT backbones with an OT alignment interface, reporting mAP gains of up to 2.5 points on DOTA and DIOR-R datasets.
ART replaces uniform attention in shallow LLM layers with local attention patterns to reduce hallucinations across multiple model architectures.
Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.
citing papers explorer
-
FASER: Fine-Grained Phase Management for Speculative Decoding in Dynamic LLM Serving
FASER delivers up to 53% higher throughput and 1.92x lower latency in dynamic LLM serving by adjusting speculative lengths per request, early pruning of rejects, and overlapping draft/verification phases via frontiers.
-
Two-dimensional early exit optimisation of LLM inference
Coordinating layer-wise and sentence-wise early exits in LLMs produces multiplicative speedups of 1.4-2.3x over single-dimension early exit on sentiment classification tasks.
-
End-to-End Dynamic Sparsity for Resource-Adaptive LLM Inference
L2A trains one LLM with input-and-budget-conditioned gates to adapt sparsity across layers, heads, and tokens, tracing the compute-accuracy frontier while staying within 0.6% of dense performance at 34% layer sparsity on tested models.
-
Tracing Computation Density in LLMs
LLM computation follows a consistent two-phase pattern: a sparse early-layer core reconstructs the head of the output distribution, with later layers and attention heads providing incremental refinements that correlate with model uncertainty.
-
Uncovering the Latent Potential of Deep Intermediate Representations
Introduces LOES, a constructive spectral method to select task-discriminative subspaces from intermediate layer embeddings, and GeoReg for enforcing simplicial class geometry during fine-tuning, with reported gains increasing with model depth across modalities.
-
The Generalization Ridge: Information Flow in Natural Language Generation
InfoRidge reveals a non-monotonic pattern in which predictive mutual information between hidden states and outputs peaks in intermediate layers before declining in final layers.
-
EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering
EASE-TTT creates a soft attention target from evidence chunks to guide query-side test-time adaptation, yielding higher macro-average scores than full-context, retrieval-only, and standard qTTT baselines on six LongBench QA tasks.
-
BMCR: Adaptive Backbone Module Composition via Reinforcement Learning for Remote Sensing Object Detection
BMCR uses RL to adaptively compose modules from CNN and ViT backbones with an OT alignment interface, reporting mAP gains of up to 2.5 points on DOTA and DIOR-R datasets.
-
ART: Attention Replacement Technique to Improve Factuality in LLMs
ART replaces uniform attention in shallow LLM layers with local attention patterns to reduce hallucinations across multiple model architectures.
-
Discovering Millions of Interpretable Features with Sparse Autoencoders
Trains and releases SAEs for Qwen3-1.7B/4B/8B models with layer-wise coverage and demonstrates causal steering of refusal via selected features.