AutoTTS discovers width-depth test-time scaling controllers through agentic search in a pre-collected trajectory environment, yielding better accuracy-cost tradeoffs than hand-designed baselines on math reasoning tasks at low cost.
hub Canonical reference
InProceed- ings of the 52nd Annual International Symposium on Computer Architecture, ISCA ’25, page 467–481, New York, NY , USA
Canonical reference. 100% of citing Pith papers cite this work as background.
27
Pith papers citing it
Background
100% of classified citations
hub tools
citation-role summary
background 6
citation-polarity summary
roles
background 6polarities
background 6representative citing papers
DASH assigns segment-level credit in reasoning traces using drift toward ground-truth answers, yielding 50.8% accuracy on AIME25 versus 45.4% for GRPO while reducing overthinking behaviors.
MARS is a margin-adversarial stopping rule for parallel LLM test-time scaling that saves 25-47% tokens while matching full-budget majority-vote accuracy by learning trace switch probabilities and applying adversarial bounds.
2-bit quantized reasoning models exhibit process failures like loops and delayed commitment that degrade end-to-end performance, but FP16 planning and loop rescue recover accuracy on MATH-500 from 17.2% to 74.2% for Qwen3-8B while retaining speed gains.
SLT selectively compresses reasoning spans via anticipation and gating, trained in three stages including RL, yielding 22.7% higher accuracy than uniform latent baselines at similar compression and 58.4% shorter chains with 2.8% accuracy drop vs explicit CoT on math benchmarks.
NPO uses a policy's own near-future checkpoint as auxiliary trajectories to maximize effective learning signal S = Q/V, improving performance from 57.88 to 63.15 on Qwen3-VL-8B-Instruct with GRPO while accelerating convergence.
RLSD mixes self-distillation for token-level policy difference magnitudes with RLVR for reliable update directions from response correctness to reach higher convergence and better training stability.
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.
Confident Decoding selects reliable intermediate layers via entropy-guided backward search to bypass final-layer alignment perturbations in LLMs, improving reasoning performance.
RKSC delivers 3.008x mean speedup over baseline and 1.66x over vLLM prefix caching for multi-branch LLM reasoning via similarity-based KV sharing and confidence-gated early exit, with 0.37% error rate.
RLVR exhibits correct-set turnover where solved problems regress during training, and a periodic review mechanism exploiting a repair-window principle improves retention and performance over baselines.
Post-training quantization increases overthinking errors in reasoning models; a logit penalty on curated overthinking markers reduces CoT length 12-23% without accuracy loss.
MSIFR stops faulty LLM generations early via staged rule-based checks, reducing token consumption 11-78% with no accuracy loss.
VISTA uses prefix resampling and a vision-aware attention score to address data imbalance and language prior bias in self-improvement training of MLLMs, yielding up to 13.66% gains on reasoning tasks.
BET reduces reasoning tokens by about 55% on average while improving performance across benchmarks by learning to short-solve easy queries, fold early on unsolvable ones, and preserve budget for hard solvable queries.
Behavior Cue Reasoning trains LLMs to emit special tokens before behaviors, enabling monitors to cut up to 50% wasted reasoning tokens and recover safe actions from 80% of unsafe traces, more than doubling success rates with no performance cost.
River-LLM enables token-level early exit in decoder-only LLMs by routing exited tokens through 4-bit quantized copies of backbone layers that share the KV cache addressing scheme, achieving 1.53–2.16× wall-clock speedup without training.
Conformal risk control with upper and lower thresholds lets LLMs adaptively stop reasoning while guaranteeing a maximum error rate and minimizing token use.
Entropy After </Think> (EAT) enables early exiting in reasoning LLMs by tracking entropy stabilization after a </think> token, cutting token use 12-22% on MATH500 and AIME2025 with no accuracy loss.
LASER reduces edge LLM serving latency by 17-38% and improves SLO satisfaction by 3-6% via load-aware adaptive early-exit thresholds and difficulty-aware budget pre-allocation, with 1% average accuracy cost.
SBBT separates Brier-score calibration gains from AUROC ranking gains in prefix-conditioned success estimation for LLM math reasoning, with structure-aware signals yielding up to +0.110 AUROC over baselines.
TRACE aggregates answer consistency and confidence trajectory over multiple reasoning steps to decide when to halt inference, reducing token usage by 25-30% while keeping accuracy within 1-2% of full reasoning.
SAT reduces reasoning tokens by up to 40% across multiple large reasoning models and benchmarks by adaptively pruning steps based on difficulty while maintaining or improving accuracy.
Reasoning LLMs aggregate social biases through stereotype repetition and irrelevant information injection in their thinking processes, and a self-review prompt mitigates this on BBQ, StereoSet, and BOLD benchmarks.
citing papers explorer
-
LLMs Improving LLMs: Agentic Discovery for Test-Time Scaling
AutoTTS discovers width-depth test-time scaling controllers through agentic search in a pre-collected trajectory environment, yielding better accuracy-cost tradeoffs than hand-designed baselines on math reasoning tasks at low cost.
-
Know When to Stop: Segment-Level Credit Assignment for Reducing Overthinking
DASH assigns segment-level credit in reasoning traces using drift toward ground-truth answers, yielding 50.8% accuracy on AIME25 versus 45.4% for GRPO while reducing overthinking behaviors.
-
MARS: Margin-Adversarial Risk-controlled Stopping for Parallel LLM Test-time Scaling
MARS is a margin-adversarial stopping rule for parallel LLM test-time scaling that saves 25-47% tokens while matching full-budget majority-vote accuracy by learning trace switch probabilities and applying adversarial bounds.
-
Extreme Low-Bit Inference in Reasoning Models: Failure Modes and Targeted Recovery
2-bit quantized reasoning models exhibit process failures like loops and delayed commitment that degrade end-to-end performance, but FP16 planning and loop rescue recover accuracy on MATH-500 from 17.2% to 74.2% for Qwen3-8B while retaining speed gains.
-
Selective Latent Thinking: Adaptive Compression of LLM Reasoning Chains
SLT selectively compresses reasoning spans via anticipation and gating, trained in three stages including RL, yielding 22.7% higher accuracy than uniform latent baselines at similar compression and 58.4% shorter chains with 2.8% accuracy drop vs explicit CoT on math benchmarks.
-
Near-Future Policy Optimization
NPO uses a policy's own near-future checkpoint as auxiliary trajectories to maximize effective learning signal S = Q/V, improving performance from 57.88 to 63.15 on Qwen3-VL-8B-Instruct with GRPO while accelerating convergence.
-
Self-Distilled RLVR
RLSD mixes self-distillation for token-level policy difference magnitudes with RLVR for reliable update directions from response correctness to reach higher convergence and better training stability.
-
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.
-
Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding
Confident Decoding selects reliable intermediate layers via entropy-guided backward search to bypass final-layer alignment perturbations in LLMs, improving reasoning performance.
-
RKSC: Reasoning-Aware KV Cache Sharing and Confident Early Exit for Multi-Step LLM Inference
RKSC delivers 3.008x mean speedup over baseline and 1.66x over vLLM prefix caching for multi-branch LLM reasoning via similarity-based KV sharing and confidence-gated early exit, with 0.37% error rate.
-
Learning to Solve, Forgetting to Retain: Correct-Set Turnover in RLVR
RLVR exhibits correct-set turnover where solved problems regress during training, and a periodic review mechanism exploiting a repair-window principle improves retention and performance over baselines.
-
Quantized Reasoning Models Think They Need to Think Longer, but They Do Not
Post-training quantization increases overthinking errors in reasoning models; a logit penalty on curated overthinking markers reduces CoT length 12-23% without accuracy loss.
-
Know When To Fold 'Em: Token-Efficient LLM Synthetic Data Generation via Multi-Stage In-Flight Rejection
MSIFR stops faulty LLM generations early via staged rule-based checks, reducing token consumption 11-78% with no accuracy loss.
-
Learn to Think: Improving Multimodal Reasoning through Vision-Aware Self-Improvement Training
VISTA uses prefix resampling and a vision-aware attention score to address data imbalance and language prior bias in self-improvement training of MLLMs, yielding up to 13.66% gains on reasoning tasks.
-
Nice Fold or Hero Call: Learning Budget-Efficient Thinking for Adaptive Reasoning
BET reduces reasoning tokens by about 55% on average while improving performance across benchmarks by learning to short-solve easy queries, fold early on unsolvable ones, and preserve budget for hard solvable queries.
-
Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight
Behavior Cue Reasoning trains LLMs to emit special tokens before behaviors, enabling monitors to cut up to 50% wasted reasoning tokens and recover safe actions from 80% of unsafe traces, more than doubling success rates with no performance cost.
-
River-LLM: Large Language Model Seamless Exit Based on KV Share
River-LLM enables token-level early exit in decoder-only LLMs by routing exited tokens through 4-bit quantized copies of backbone layers that share the KV cache addressing scheme, achieving 1.53–2.16× wall-clock speedup without training.
-
Conformal Thinking: Risk Control for Reasoning on a Compute Budget
Conformal risk control with upper and lower thresholds lets LLMs adaptively stop reasoning while guaranteeing a maximum error rate and minimizing token use.
-
Entropy After </Think> for reasoning model early exiting
Entropy After </Think> (EAT) enables early exiting in reasoning LLMs by tracking entropy stabilization after a </think> token, cutting token use 12-22% on MATH500 and AIME2025 with no accuracy loss.
-
LASER: Load-Aware Serving with Early-Exit for Reasoning LLMs at the Edge
LASER reduces edge LLM serving latency by 17-38% and improves SLO satisfaction by 3-6% via load-aware adaptive early-exit thresholds and difficulty-aware budget pre-allocation, with 1% average accuracy cost.
-
Prefix-Safe Bayesian Belief Tracking for LLM Reasoning Reliability:Separating Calibration from Ranking
SBBT separates Brier-score calibration gains from AUROC ranking gains in prefix-conditioned success estimation for LLM math reasoning, with structure-aware signals yielding up to +0.110 AUROC over baselines.
-
Efficient Test-Time Scaling via Temporal Reasoning Aggregation
TRACE aggregates answer consistency and confidence trajectory over multiple reasoning steps to decide when to halt inference, reducing token usage by 25-30% while keeping accuracy within 1-2% of full reasoning.
-
SAT: Balancing Reasoning Accuracy and Efficiency with Stepwise Adaptive Thinking
SAT reduces reasoning tokens by up to 40% across multiple large reasoning models and benchmarks by adaptively pruning steps based on difficulty while maintaining or improving accuracy.
-
Investigating Thinking Behaviours of Reasoning-Based Language Models for Social Bias Mitigation
Reasoning LLMs aggregate social biases through stereotype repetition and irrelevant information injection in their thinking processes, and a self-review prompt mitigates this on BBQ, StereoSet, and BOLD benchmarks.
-
DRP: Distilled Reasoning Pruning with Skill-aware Step Decomposition for Efficient Large Reasoning Models
DRP combines teacher-guided pruning of chain-of-thought steps with distillation to cut token usage in reasoning models on GSM8K and AIME while maintaining or improving accuracy.
-
EasyVideoR1: Easier RL for Video Understanding
EasyVideoR1 delivers an optimized RL pipeline for video understanding in large vision-language models, achieving 1.47x throughput gains and aligned results on 22 benchmarks.
- Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thought