Introduces the Generalization Spectrum evaluation framework to track per-example generalization across transfer distances in competitive programming tasks.
hub
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
31 Pith papers cite this work, alongside 1 external citations. Polarity classification is still indexing.
31
Pith papers citing it
1
external citations · Pith
abstract
Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments on Qwen3-14B models using math-only data but different tuning methods. We find that reinforcement learning (RL)-tuned models generalize well across domains, while supervised fine-tuning (SFT)-tuned models often forget general capabilities. Latent-space representation and token-space distribution shift analyses reveal that SFT induces substantial representation and output drift, while RL preserves general-domain structure. Our results suggest a need to rethink standard post-training recipes, particularly the reliance on SFT-distilled data for advancing reasoning models.
hub tools
citation-role summary
background 2
method 2
citation-polarity summary
representative citing papers
TAC is a bandit curriculum for multi-domain RLVR that prioritizes domains whose gradient updates align with and benefit other domains, yielding up to 2.8-point macro accuracy gains over learnability-only baselines on Qwen3-1.7B and Llama3.2-3B.
OmniOPD replaces token-level logit matching in on-policy distillation with Monte Carlo chunk-level semantic verification and a peak-entropy scheduler.
Temporal scheduling of credit allocation criteria over RLVR training, using trajectory percentiles to target heterogeneous behaviors, yields more stable policy entropy and better reasoning benchmark results than static allocation.
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
A single LLM improves its own reasoning by self-distilling from privileged verified traces as teacher to its question-only student policy, outperforming off-policy distillation and RL on math benchmarks with better token efficiency.
SFT supplies entangled compositional traces of atomic skills and routing modules; RL identifies those modules and enables recombination on novel compositions outside the SFT support.
PartRep selects high-NLL tokens via a lightweight early-exit gate for partial prompt repetition, retaining most full-repetition gains at 59.4% KV cache and 79% prefill FLOPs on eight benchmarks.
Introduces RSI metric and RSI-S filtering method for adaptive token selection in RLVR, reporting 2-3 point gains over GRPO on AIME/AMC benchmarks.
The LLM-as-Environment-Engineer framework lets the policy model redesign its own RL environments on the new MAPF-FrozenLake testbed, outperforming larger models and fixed baselines with Qwen3-4B.
CoT SFT disrupts long-range routing in hybrid models via changes to W_Q and W_K; QK-Restore restores pre-SFT projections to recover NIAH performance.
Modular AI systems assembled from contributed small models outperform monolithic LLMs by up to 15.4% on 15 tasks including reasoning and factuality while showing emergent problem-solving and benefits from contributor diversity.
Later-domain RL training harms earlier domains via second-order damage concentrated in a low-dimensional shared conflict subspace; brief domain refresh contracts this component to enable selective recovery.
DelTA estimates token coefficients to amplify discriminative directions in token-gradient vectors, reweighting the RLVR surrogate to produce more contrastive side-wise centroids and yielding 3.26 and 2.62 point gains on math benchmarks for 8B and 14B Qwen3 models.
SSOPD converts intra-group correct-wrong contrast into process supervision by distilling a teacher distribution from the shortest correct completion into prefixes of the longest wrong completion, improving GRPO on AIME and HMMT benchmarks.
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
RL generalizes better than SFT by preserving and slowly evolving a compact set of task-agnostic features from the base model rather than introducing many specialized ones.
HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.
Recovering an orthogonal basis from model activations yields a model-native skill characterization that improves reasoning Pass@1 by up to 41% via targeted data selection and supports inference steering, outperforming human-characterized alternatives.
ETS performs training-free RL alignment for language models by energy-guided test-time scaling with Monte Carlo energy estimation and importance sampling acceleration.
ARS shapes reasoning trace representations by clustering states that produce consistent answers and separating those that produce inconsistent ones via latent perturbations, improving plug-and-play hallucination detection without human annotations.
SFT and RL cannot be decoupled in LLM post-training because each step increases the loss or lowers the reward of the prior step under KL and PL analyses.
Finetuning VLMs on perception tasks produces positive and negative transfers that can be mapped with a new normalized metric called Perfection Gap Factor across 13 tasks and three models.
Curating 25K natural-instruction examples for RLVR yields a 64.4pp relative gain on BBEH for Qwen3-0.6B and generalizes across scales and model families.
citing papers explorer
-
The Generalization Spectrum: A Chromatographic Approach to Evaluating Learning Algorithms
Introduces the Generalization Spectrum evaluation framework to track per-example generalization across transfer distances in competitive programming tasks.
-
Transferability for General Reasoning: An Automated Curriculum for Multi-Domain RLVR
TAC is a bandit curriculum for multi-domain RLVR that prioritizes domains whose gradient updates align with and benefit other domains, yielding up to 2.8-point macro accuracy gains over learnability-only baselines on Qwen3-1.7B and Llama3.2-3B.
-
OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification
OmniOPD replaces token-level logit matching in on-policy distillation with Monte Carlo chunk-level semantic verification and a peak-entropy scheduler.
-
Not only where, But when: Temporal Scheduling for RLVR
Temporal scheduling of credit allocation criteria over RLVR training, using trajectory percentiles to target heterogeneous behaviors, yields more stable policy entropy and better reasoning benchmark results than static allocation.
-
Deep Reasoning in General Purpose Agents via Structured Meta-Cognition
DOLORES, an agent using a formal language for meta-reasoning to construct adaptive scaffolds on the fly, outperforms prior scaffolding methods by 24.8% on average across four hard benchmarks and multiple model sizes.
-
Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models
A single LLM improves its own reasoning by self-distilling from privileged verified traces as teacher to its question-only student policy, outperforming off-policy distillation and RL on math benchmarks with better token efficiency.
-
From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning
SFT supplies entangled compositional traces of atomic skills and routing modules; RL identifies those modules and enables recombination on novel compositions outside the SFT support.
-
PARTREP: Learning What to Repeat for Decoder-only LLMs
PartRep selects high-NLL tokens via a lightweight early-exit gate for partial prompt repetition, retaining most full-repetition gains at 59.4% KV cache and 79% prefill FLOPs on eight benchmarks.
-
Which Tokens Matter? Adaptive Token Selection for RLVR with the Relative Surprisal Index
Introduces RSI metric and RSI-S filtering method for adaptive token selection in RLVR, reporting 2-3 point gains over GRPO on AIME/AMC benchmarks.
-
From Trainee to Trainer: LLM-Designed Training Environment for RL with Multi-Agent Reasoning
The LLM-as-Environment-Engineer framework lets the policy model redesign its own RL environments on the new MAPF-FrozenLake testbed, outperforming larger models and fixed baselines with Qwen3-4B.
-
Attention Amnesia in Hybrid LLMs: When CoT Fine-Tuning Breaks Long-Range Recall, and How to Fix It
CoT SFT disrupts long-range routing in hybrid models via changes to W_Q and W_K; QK-Restore restores pre-SFT projections to recover NIAH performance.
-
Scaling Participation in Modular AI Systems
Modular AI systems assembled from contributed small models outperform monolithic LLMs by up to 15.4% on 15 tasks including reasoning and factuality while showing emergent problem-solving and benefits from contributor diversity.
-
A Local Perturbation Theory for Cross-Domain Interference and Recovery in Multi-Domain RL
Later-domain RL training harms earlier domains via second-order damage concentrated in a low-dimensional shared conflict subspace; brief domain refresh contracts this component to enable selective recovery.
-
DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards
DelTA estimates token coefficients to amplify discriminative directions in token-gradient vectors, reweighting the RLVR surrogate to produce more contrastive side-wise centroids and yielding 3.26 and 2.62 point gains on math benchmarks for 8B and 14B Qwen3 models.
-
Self-Supervised On-Policy Distillation for Reasoning Language Models
SSOPD converts intra-group correct-wrong contrast into process supervision by distilling a teacher distribution from the shortest correct completion into prefixes of the longest wrong completion, improving GRPO on AIME and HMMT benchmarks.
-
Rotation-Preserving Supervised Fine-Tuning
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
-
Why Does Reinforcement Learning Generalize? A Feature-Level Mechanistic Study of Post-Training in Large Language Models
RL generalizes better than SFT by preserving and slowly evolving a compact set of task-agnostic features from the base model rather than introducing many specialized ones.
-
HEALing Entropy Collapse: Enhancing Exploration in Few-Shot RLVR via Hybrid-Domain Entropy Dynamics Alignment
HEAL mitigates entropy collapse in few-shot RLVR by selectively adding general-domain data and aligning trajectory-level entropy dynamics, matching full-shot performance with 32 target samples.
-
Characterizing Model-Native Skills
Recovering an orthogonal basis from model activations yields a model-native skill characterization that improves reasoning Pass@1 by up to 41% via targeted data selection and supports inference steering, outperforming human-characterized alternatives.
-
ETS: Energy-Guided Test-Time Scaling for Training-Free RL Alignment
ETS performs training-free RL alignment for language models by energy-guided test-time scaling with Monte Carlo energy estimation and importance sampling acceleration.
-
Harnessing Reasoning Trajectories for Hallucination Detection via Answer-agreement Representation Shaping
ARS shapes reasoning trace representations by clustering states that produce consistent answers and separating those that produce inconsistent ones via latent perturbations, improving plug-and-play hallucination detection without human annotations.
-
On the Non-decoupling of Supervised Fine-tuning and Reinforcement Learning in Post-training
SFT and RL cannot be decoupled in LLM post-training because each step increases the loss or lowers the reward of the prior step under KL and PL analyses.
-
Understanding Task Transfer in Vision-Language Models
Finetuning VLMs on perception tasks produces positive and negative transfers that can be mapped with a new normalized metric called Perfection Gap Factor across 13 tasks and three models.
-
SUPERNOVA: Eliciting General Reasoning in LLMs with Reinforcement Learning on Natural Instructions
Curating 25K natural-instruction examples for RLVR yields a 64.4pp relative gain on BBEH for Qwen3-0.6B and generalizes across scales and model families.
-
M2A: Synergizing Mathematical and Agentic Reasoning in Large Language Models
M2A uses null-space model merging to combine mathematical and agentic reasoning in LLMs, raising SWE-Bench Verified performance from 44.0% to 51.2% on Qwen3-8B without retraining.
-
Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning
Sequential SFT followed by RL, guided by the Plasticity-Ceiling Framework, achieves higher performance ceilings in LLM mathematical reasoning than synchronized methods by optimizing data scale and transition timing.
-
Proximal Supervised Fine-Tuning
PSFT modifies supervised fine-tuning by incorporating trust-region ideas from RL to constrain policy changes, yielding better out-of-domain generalization in math and human-value tasks without entropy collapse.
-
Sample-efficient LLM Optimization with Reset Replay
LoRR augments preference optimization methods like DPO with high-replay training, periodic resets to initial data/policy, and a hybrid objective to improve sample efficiency and reduce primacy bias on math and reasoning tasks.
-
Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
The paper unifies perspectives on Long CoT in reasoning LLMs by introducing a taxonomy, detailing characteristics of deep reasoning and reflection, and discussing emergence phenomena and future directions.
-
A Survey of Reinforcement Learning for Large Reasoning Models
A survey compiling RL methods, challenges, data resources, and applications for enhancing reasoning in large language models and large reasoning models since DeepSeek-R1.
- Token Buncher: Shielding LLMs from Harmful Reinforcement Learning Fine-Tuning