LLM Post-Training

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

29 papers in the last four weeks, up 81% on the four weeks before. 0.3% of all new papers.

Jul 13Week of Sep 28

Latest papers 175

Oct 7, 2026cs.AI

Successive Training Stages and Large Language Model Persuasion: Effects of Misalignment, Supervised Fine-Tuning, and Preference Optimization

Large language models (LLMs) can be tuned to influence human attitudes, yet the respective contributions of successive post-training stages remain un-clear. This study examines how three successive training stages affect LLM persuasiveness: (1) misalignment through supervised fine-tuning (SFT) on conspiracy data, (2) additional persuasive SFT on argumentative data, and (3) Identity Preference Optimization (IPO), a preference-optimization method. A total of 835 participants recruited on Prolific were randomly assigned to five between-subject conditions (neutral text, conspiracy-trained model, persuasion-trained model, preference-optimized model, and GPT-4) and were exposed to texts on 10 divisive political issues, personalized from their individual profiles in all model conditions. Attitude change was measured as the difference between pre- and post-exposure positions on continuous Likert scales and analyzed with an analysis of covariance (ANCOVA). A significant condition x baseline-attitude interaction, F (4, 825) = 5.33, p < .001, indicated that training effects depended on participants' initial attitudes. Persuasive SFT produced greater attitude change than conspiracy training alone, d = 0.30, whereas IPO provided no additional benefit, d = 0.03, and GPT-4 did not differ from neutral text, d = --0.01. These results show that targeted supervised training on persuasive data increases LLM persuasiveness, whereas preference optimization yields no significant gains beyond it.

Antony Dalmiere, Pascal Marchand, Guillaume Auriol +1

LAAS-TRUST · LAAS-TRUST, INSA Toulouse · LAAS-TSF, LAAS

Persuasion in Language Models LLM Post-Training Supervised Fine-Tuning Preference Optimization

Oct 7, 2026cs.CL

SAPD: Step-Aligned Privileged Distillation

On-policy post-training can improve large language models by learning from their own trajectories, but requires costly rollout generation. We ask whether fixed demonstrations can support competitive off-policy learning through better supervision. Our premise is that their usefulness depends not only on the training trajectories, but also on whether supervision provides informative preferences among continuations and connects this guidance to the reasoning decision being learned. We introduce Step-Aligned Privileged Distillation (SAPD), a rollout-free self-distillation method that turns demonstrations into step-aligned distributional supervision. Its key insight is to use the known progression of a reference solution to associate each reasoning transition with targeted privileged guidance, rather than treating the solution as undifferentiated context. On mathematical reasoning benchmarks, SAPD outperforms supervised fine-tuning and label smoothing on average while remaining competitive with on-policy reinforcement learning and self-distillation. Analyses support both the value of context-dependent distributional guidance and the benefit of aligning privileged information with the current step. SAPD also largely preserves out-of-domain coding performance and achieves approximately 2x training-loop speedups over the on-policy baselines. These findings suggest that carefully constructed supervision can make fully off-policy post-training a competitive and computationally efficient alternative. Our code is available at https://github.com/Miaow-Lab/SAPD.

Tianle Wang, Jiayu Liu, Ruizhi Zhao +1

Department of Data Science, City University of Hong Kong · Hong Kong Institute of AI for Science, City University of Hong Kong · Shenzhen University

LLM Post-Training Language Model Distillation Knowledge Distillation LLM Reasoning Self-Distillation

Oct 7, 2026cs.LG

When Rank Rises as LLMs Degrade

Post-training adapts language models in non-stationary environments. Practitioners monitor representation health with RankMe and related spectral statistics, often assuming that rank falls when representations degrade. We show that this assumption is unsafe for LLM post-training. In a controlled study of Qwen3-0.6B with four degradation modes and three seeds, data duplication worsens held-out loss by 75% relative to healthy while increasing both original and centred RankMe; the latter changes by 13.5 pooled standard deviations. Covariance effective rank rises to nearly twice its healthy value. This failure is spectral dispersion rather than collapse, so a one-sided monitor rates the worst checkpoint as the healthiest. By contrast, a learning-rate misconfiguration lowers centred RankMe and k95, while uncentred RankMe is inconsistent across seeds. Direction is therefore a property of the regime-statistic pair and cannot be fixed by recalibration alone. We also distinguish two often-conflated statistics: RankMe normalises singular values, whereas covariance effective rank normalises eigenvalues. On raw intermediate-layer states in the pretrained model, massive activations pin the latter near 1 out of dimension d while RankMe retains usable range. We then test a two-sided, multichannel sequential monitor with separate calibration and test data. In a pre-registered shared-prefix, leave-one-seed-out evaluation, it detects all three damage regimes in every fold 10 to 60 steps after the fork and separates dispersion from downward-rank damage by firing direction. However, it never precedes held-out probe loss, and calibration with two seeds produces false alarms on the held-out healthy seed. Spectral monitoring can diagnose failure regimes, but it does not warn earlier than held-out loss, and validity claims require held-out healthy data.

Zhaohui Geoffrey Wang

University of Southern California

LLM Post-Training Representation Geometry in Language Models

Oct 6, 2026cs.AI

CM-DPO: Constraint-Margin Direct Preference Optimization for LLM Planning

Direct Preference Optimization (DPO) treats all constraint violations equally: a

1 budget overshoot and a

1,000 overshoot induce the same training signal. It is also susceptible to length and style bias when preference pairs come from different model families. We introduce Constraint-Margin DPO (CM-DPO), which replaces DPO's binary preference signal with a continuous margin derived from a deterministic symbolic verifier and scaled by violation severity. Hard and soft constraints are separated through a lexicographic objective, ensuring hard constraints are never traded off against preferences. To supply CM-DPO with bias-reduced training pairs, we generate preference data through procedurally generated constraint profiles (DCCG) and minimal-edit distillation from a reasoning teacher (RT-MED), within a framework we call SynPlan-R. On TravelPlanner, NaturalPlan, and out-of-distribution PlanBench, an 8B model fine-tuned with CM-DPO achieves 89.2% pass rate and 93.4% solve rate, matching multi-agent systems at 13x lower latency while outperforming GPT-4o on unseen Blocksworld by 9.2 points.

Rabimba Karanjai, Qun Gu, Hemanth Hegadehalli Madhavarao +8

PayPal AI Lab

Direct Preference Optimization LLM Planning LLM Post-Training

Oct 5, 2026cs.LG

Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates

Safety auditing of post-trained large language models typically relies on model behavior, requiring model execution and depending on the coverage of available evaluations. This work asks a different question: Do the target behaviors optimized during supervised fine-tuning (SFT) leave readable evidence directly in checkpoint updates? We find that harmful-compliance SFT induces a continuous, objective-dependent ordering in checkpoint-update space. Using a reference geometry defined by pure harmful-compliance, safety-targeted, and benign-utility SFT, we find that a checkpoint-level coordinate s_H tracks controlled harmful-objective composition with Spearman correlations of 0.986-0.992 across four 7-8B backbones, with the same ordering persisting at larger model scales. Matched compliance-versus-refusal controls show that this checkpoint trace reflects the SFT objective rather than harmful-input exposure, while additional controls rule out simple explanations based on harmful-example count or generic training intensity. Building on this structure, we introduce TRACE, a weights-only auditing method that localizes an unknown checkpoint update relative to frozen harmful and non-harmful reference prototypes and converts this geometry into a continuous harmful-objective score. TRACE requires neither model queries nor access to the unknown SFT data, and can be evaluated directly from checkpoint updates. Across distribution shifts, unseen data, different SFT configurations, partial checkpoint access, and LoRA/full-parameter fine-tuning, the trace remains stable and is positively associated with independently measured attack success rates. TRACE remains informative even at low harmful-objective proportions, providing a complementary auditing signal when behavioral evaluation is unavailable or incomplete. Code is available at https://anonymous.4open.science/r/Code4TRACE-54D3.

Ziqun Bao, Xinyu Zhang, Yuchen Shao +1

SEI, East China Normal University

LLM Auditing LLM Safety LLM Post-Training

Oct 5, 2026cs.CL

The Assistance Dilemma: Learning to Teach via Multi-Turn Reinforcement Learning

Large language models (LLMs) trained to answer questions are natively poor at teaching. Reinforcement Learning (RL) against a simulated student is a promising approach to improve their pedagogy, but existing RL-trained tutors reward the student's success on the tutored problem with the tutor's words still in context. The reward is then easiest to raise by telling the student the answer, and a tuned penalty is needed to reduce telling. Drawing on learning sciences, we introduce a masked near-transfer post-test: the student is tested on an unseen variant of the tutored problem with the tutor's utterances masked, so the reward can rise only through what the student wrote in its own turns. This discourages cognitive offloading by the student and allows the continuous penalty to be replaced by two binary reward gates (factual correctness of tutor response, no solution handover). A leave-one-out ablation shows that the learning-gain reward on its own does not separate teaching from telling: the gates reduce solution handover while the near-transfer post-test improves out-of-domain transfer. Using these reward designs we develop Eduardo, a multi-turn RL recipe for training LLM tutors, and use it to train 4B, 9B, 14B and 27B models from two distinct LLM architectures. Our post-trained Eduardo-27B model matches Gemini-3.1-Pro on MathTutorBench and Claude Opus 4.8 on TutorMoments at 2.4-6.2x fewer thinking tokens than frontier models, which matters for interactive tutoring. Without being named in the reward, the model more than doubles its use of the push-for-justification teacher move while support fading (e.g., assigning independent work), whose payoff lies beyond a single-problem dialog episode, is trained out. We open-source our training environment, an 8,671-problem near-transfer dataset, and trained models for further development.

Jakub Macina, Manu Kapur, Mrinmaya Sachan

ETH Zurich

RL for Language Models Intelligent Tutoring Systems LLM Post-Training

Oct 5, 2026cs.AI

ImproveAnyTask: An Autonomous Post-Training Harness for Iterative Model Self-Improvement

Adapting general-purpose large language models to specific tasks requires substantial human effort in designing data and training strategies. Sustaining improvement is especially challenging because model updates change the error distribution, requiring strategies to be continually refined. We introduce ImproveAnyTask, an autonomous post-training harness that improves task performance under a limited compute budget. Drawing inspiration from gradient-based parameter optimization, the harness organizes adaptation into error attribution, update-direction selection, and executable model updates. It combines metric-level and case-level analysis to identify a focal problem, then investigates research-backed strategies and compares their reported gains and reproduction difficulty. The selected strategy is translated into training data and a training configuration, with small-scale execution checks preceding full post-training. Subsequent evaluation guides model selection and further adaptation, while validated strategies and scripts are retained for reuse. Across 11 tasks, ImproveAnyTask achieves mean gains of 18.29 and 11.97 percentage points on the Base and Instruct models, respectively, with a maximum gain of 41.96 points, under a 24-hour budget with resources equivalent to eight H20 GPUs.

Xingbo Yao, Xiaoman Wang, Zhengwu Lei +11

HKUST(GZ) · Xiaohongshu Inc. · ECNU +3

LLM Fine-Tuning LLM Post-Training Language Model Post-Training Language Model Self-Improvement

Oct 5, 2026cs.LG

Dynamic Minimax Regret Optimization for Robust LLM Post-Training

Modern LLM training increasingly relies on heterogeneous data sources spanning different domains, tasks, preference distributions, and difficulty levels. We study dynamic minimax regret for group-distributionally robust LLM post-training under instantaneous mini-batch-only bandit feedback. The framework views the training as a two-player sampler-optimizer process: a sampler adaptively selects among data sources using bandit feedback, while an optimizer updates the model parameters using stochastic gradients from the selected source. We focus on the practically restrictive setting where source losses evolve with model training but historical data are not re-evaluated, requiring the sampler to track instantaneous worst-sources from stale partial feedback. We propose DUCB-OGD, a simple and scalable algorithm that couples a Discounted Upper-Confidence-Bound sampler with an Online Gradient Descent optimizer. The sampler maintains exponential moving average loss estimates and confidence radii based on discounted effective sample sizes, avoiding costly re-evaluation of past data or intrusive changes to standard training pipelines. For

K

data sources and

T

training steps, we prove that DUCB-OGD achieves a dynamic minimax regret of

\tilde{O}(K^{1/4}T^{3/4})

, which is optimal up to logarithmic factors for the undiscounted objective under our feedback model. Extensive experiments across supervised fine-tuning, preference optimization, and reinforcement learning show that DUCB-OGD integrates seamlessly into modern LLM training pipelines and improves worst-group robustness with negligible computational overhead compared with standard sampling baselines.

Chengbo Zang, Haoyu Dong, Mehmet Kerem Turkcan +3

Department of Electrical Engineering, Columbia University · Department of Computer Science & Engineering, University of California, Santa Cruz · Department of Civil Engineering, Columbia University

Dynamic Regret Minimization Distributionally Robust Optimization Robust Multi-Armed Bandits LLM Post-Training LLM Training

Oct 4, 2026cs.AI

The Functional Structure of Post-Compression Recovery in Low-Rank LLMs

Different low-rank compression methods can produce compressed LLMs that respond differently to the same post-compression recovery procedure, and relative advantages observed between methods at the compression endpoint may shrink, grow, or even reverse after recovery. We ask whether this recovery heterogeneity reflects functional structure beyond scalar loss evolution, and how that structure evolves throughout recovery. Our results establish that this heterogeneity reflects a reproducible compression-induced functional structure, which we formalize as recovery pressure. To characterize this structure consistently throughout recovery, we develop a standardized functional characterization within each backbone that is applicable across heterogeneous low-rank methods. The primary backward characterization reveals reproducible module-wise structure across independent probes, while a complementary forward-only characterization recovers related structure without loss or backpropagation. We further find that recovery pressure measured at the endpoint is associated with subsequent recovery response; during recovery, its module-wise structure is reorganized non-uniformly, and localized updates induce distributed responses beyond directly updated modules. Further evidence indicates that tracking this evolving structure provides a complementary functional view of recovery progress alongside scalar loss.

Zishan Shao, Liang Tian, Georgiy Zemlevskiy +10

Duke University · Wake Forest University · Carnegie Mellon University +2

LLM Compression Low-Rank Compression LLM Post-Training

Oct 1, 2026cs.LG

The Missing Primitive: Diagnosing and Repairing Mathematical Reasoning in Large Language Models

While Large Language Models (LLMs) have demonstrated striking capabilities on frontier mathematical problems, it remains unclear whether they possess the structural mathematical understanding underlying their solutions. In this paper, we take a first step toward systematically studying mathematical understanding in LLMs, from diagnosing its distinct capabilities to leveraging these findings to improve post-training. First, we introduce the notion of Mathematical Primitive to probe structural mathematical understanding and propose \hlei{}, a novel benchmark that evaluates mathematical reasoning along four distinct dimensions: Discovery, Generation, Digestion, and Execution. Second, our systematic diagnosis shows that solution accuracy masks distinct capability profiles, primitives unlock substantial latent execution capacity, and Discovery is the dominant bottleneck in mathematical reasoning. Our post-training analysis further shows that discovery-limited failures are particularly amenable to repair. Finally, building on these findings, we introduce \abs{}, a primitive-privileged self-distillation framework that selectively transfers primitive-guided reasoning into the student model. Extensive experiments demonstrate that \abs{} consistently improves mathematical reasoning over baselines across model scales and challenging benchmarks.

Shuo Xing, Zilin Dai, Chengyuan Qian +7

Texas A&M University · Harvard University · Vanderbilt University +3

Mathematical Reasoning Benchmarks Self-Distillation LLM Post-Training LLM Mathematical Reasoning

Sep 30, 2026cs.AI

Agent Error Dataset: Scaling 50,000 Error--Diagnosis Pairs for Failure Analysis and Error-Aware Post-Training

An unsuccessful LLM agent rollout contains more information than its final reward: the observations available to the agent, the actions it chose, and the environment's responses. Reusing this experience for learning requires identifying a decision to revise and testing a concrete alternative. We introduce the Agent Error Dataset (AED), comprising 50,228 error-diagnosis pairs from 9,961 source tasks across 33 environments, 19 harness families, and 23 policy models in text-based agent systems. We retain source traces and execution metadata to support cross-setting failure analysis and re-diagnosis without repeating the original rollout. Our five-stage Agentic Error-to-Training (AET) pipeline collects natural failures, generates diagnoses and proposed corrections, and checks them against recorded evidence. Where replay is supported, we compare corrections with original-action retries from the same checkpoint under matched execution settings. We then construct separate training views for diagnosis and actor recovery. Across 3,062 matched replay pairs, first-proposal corrections raise verifier pass rates from 18.4% to 51.1%, a gain of 32.7 percentage points. Using a separately frozen diagnosis release, full-diagnosis fine-tuning on 1,656 source tasks raises Qwen3-8B's exact-step agreement with internal teacher labels from 47.2% to 63.6%, averaged over three seeds on a 943-case holdout. The strongest prompted reference in this comparison scores 54.7%, and mean agreement improves at each of four increasing training-set sizes. In a single-seed comparison of actor-training recipes, action-only repair training scores 6.67 percentage points higher on WebShop-lite than success-only training.

Kunlun Zhu, Xuyan Ye, Yibo Li +3

Apodex

Agent Failure Analysis LLM Agent Training LLM Post-Training LLM Agent Reliability

Sep 30, 2026cs.LG

IrekoGPT: Turning Structured Pruning into Post-Hoc Slimmable LLMs

We introduce IrekoGPT, a post-hoc method for converting pretrained LLMs into slimmable models whose width can be adjusted at inference time. Building on SliceGPT, we retain its projection matrices without pruning them, allowing a single model to expose nested subnetworks at different widths. We improve robustness by calibrating each layer across multiple compression ratios, and correct downstream linear layers through gradient-free ridge regression. Across Llama and Qwen models, preliminary results show improvements over naive PCA-based slimming, with the largest gains at high compression. Code is available at https://github.com/aimagelab/IrekoGPT

Pietro Moriello, Pietro Buzzega, Angelo Porrello +1

AImageLab, University of Modena and Reggio Emilia, Italy

LLM Compression LLM Inference Acceleration LLM Post-Training

Sep 30, 2026cs.LG

From Search to Signal: Online Post-Training in Automatic Heuristic Design

Large language model (LLM)-based automatic heuristic design (AHD) iteratively proposes and refines heuristics, pairing design rationales with executable code. Task-specific evaluators assess programs; execution outcomes and performance scores guide search. Many AHD systems keep the generator frozen; EvoTune and Co-Evolution of Algorithms and Language Model (CALM) instead update it from evaluated candidates. When such outcomes drive reinforcement learning with verifiable rewards (RLVR), they create a search-coupled loop: the evaluated candidate stream supplies both search-state updates and training signals for the model that generates future candidates. Yet validity and performance do not uniquely determine useful model updates; converting them into learning signals must account for the prompt and evolving search state that produced each candidate. We formulate online post-training of small open-weight LLMs in AHD as context-dependent signal construction and develop alternative mappings from program validity, task performance, and generation context to update signals. Using shared evaluated rollouts and matched update budgets, controlled experiments across AHD tasks and model families compare these mappings with online post-training baselines, testing their effects on validity, performance among valid proposals, and the yield of valid proposals that improve under contextual comparisons. Complementary checkpoint, frozen-search, and live-system evaluations assess whether proposal-level gains appear in updated checkpoint behavior and subsequent search, rather than arising solely from accumulated search state. A resource-matched comparison under pre-specified cost accounting tests whether online updating adds value beyond additional search with a frozen generator. Together, this design avoids treating end-to-end search gains alone as evidence of stronger heuristic-design capabilities.

Yilun Yuan, Tianyu Zhou, Zhenzhou Tang

Wenzhou University

Automated Heuristic Design Reinforcement Learning with Verifiable Rewards LLM Post-Training Language Model Post-Training

Sep 29, 2026cs.AI

Aligned Data Can Induce Misalignment via Context Confusion

Large language models (LLMs) are frequently updated for various use cases, where filtering out misaligned training samples is a common practice for preventing post-update misalignment. However, alignment is inherently context-dependent: a recommendation that is aligned in one context may be inappropriate in another. For example, in response to the question "What should a researcher do with the research data?", recommending that the researcher preserve the data for reproducibility is aligned. In contrast, recommending data saving in response to "What should a mobile-app developer do with users' sensitive data?" may be inappropriate from a privacy perspective. Starting from this observation, we identify a post-training phenomenon where aligned training induces misaligned behavior in other contexts. We call this phenomenon context confusion. We demonstrate context confusion across three domains: (1) Gender Equality, (2) Privacy, and (3) Physical Safety. We further show that context confusion causes narrow misalignment, in contrast to emergent misalignment, and is not effectively reduced by injecting general alignment data, but can be substantially reduced by including targeted alignment data for the misaligned domain or providing in-context learning examples during inference. Lastly, we provide a mechanistic explanation of context confusion. We observe that queries from different domains can undergo similar representational shifts during the fine-tuning. Consequently, a query from a different domain may activate the same behavioral feature learned during fine-tuning, which causes the behavior to transfer to a context where it is misaligned. Based on our findings, we argue that it is difficult to predict the alignment state of a model after training by inspecting the training data alone, which highlights the importance of comprehensive post-training alignment evaluations.

Yavuz Bakman, Duygu Nur Yaldiz, Baris Askin +4

University of Southern California · Carnegie Mellon University · Amazon +1

LLM Alignment LLM Post-Training Inference-Time Language Model Alignment

Sep 29, 2026cs.LG

Beyond Compression: Diagnosing How Post-Training Changes Mathematical Reasoning

Post-training is central to mathematical reasoning in modern large language models (LLMs), but endpoint pass@1 alone underidentifies what has changed. Gains may reflect newly reachable solutions, cheaper sampling of latent solutions, surface robustness, or memorisation. We compare three post-training paths under a common diagnostic readout: our sufficiently trained off-policy distillation trajectories, released Qwen3 off-policy-plus-on-policy distillation endpoints, and a released DeepSeek-Math endpoint trained with Group Relative Policy Optimisation (GRPO). Our probe uses cross-surface pass@K over verbatim prompts, paraphrases, numerical isomorphisms, and translations, plus consistency, distribution-shape, and verified supervised-fine-tuning (SFT) membership analyses. We find two regimes. On easier AMC problems, large-K ceilings are near saturation, so post-training mainly compresses sample cost. On harder AIME problems, post-training expands the large-K ceiling over the base model: sufficient off-policy distillation already raises this ceiling, Qwen3 released endpoints raise it further, and DeepSeek-Math GRPO does not dominate sufficient off-policy distillation at large K. English-dominant distillation improves non-English reasoning but preserves language-tier gaps. A controlled-overfit audit finds limited sensitivity in current SFT-membership probes. Compression is one regime of post-training, not a universal explanation.

Hongyang Li, Yiming Zhu, Xiao Li +3

University of Luxembourg · Seafill Open-Source Community · Université Paris-Saclay

Language Model Distillation RL Post-Training LLM Post-Training LLM Mathematical Reasoning

Sep 29, 2026cs.AI

WEFT: Scaling Tool-Use Post-Training for General-Purpose Agents

Recent efforts to scale tool-use post-training have largely centered on the synthesis of executable environments, which constitute only one component of a broader agentic interaction system comprising the environment, task, agent harness, and evaluator. Scaling environments in isolation, however, does not guarantee commensurate gains in model performance, because reliable learning signals depend on coherent interactions among all components of the agentic interaction system. To address this problem, we introduce WEFT (Whole-system Evolution For Tool-use Post-training), which couples scalable agentic interaction system construction, execution-driven self-evolution, and stable post-training. WEFT scales agentic interaction system construction across environment breadth, task complexity, and interaction diversity. Execution-driven self-evolution iteratively uses execution traces and state evidence to attribute failures and revise the responsible components, with fresh rollouts evaluating the changes and providing evidence for subsequent evolution rounds. For stable post-training at scale, WEFT addresses both optimization and execution reliability: prefix-preserving sampling retains verified progress and atomic-turn credit assignment localizes learning signals, while MegaMCP maintains isolated, recoverable state across concurrent rollouts over shared tool services. Extensive experiments across various models and benchmarks demonstrate the effectiveness of WEFT for tool-use post-training. WEFT-8B and WEFT-14B outperform all evaluated matched-size environment-scaling baselines on BFCL V4,

τ^2

-Bench, and Claw-Eval. In particular, WEFT-14B improves over Agent-World-14B by 6.41, 2.23, and 12.27 percentage points. WEFT-35B-A3B further extends these gains to more challenging long-horizon workflow benchmarks, including Toolathlon-Verified and AutomationBench.

Bo Mao, Hang He, Linting Wang +17

East China Normal University · Shanghai Innovation Institute · Fudan University +2

LLM Agent Self-Improvement LLM Post-Training RL for Tool Use

Sep 28, 2026cs.LG

Frontier Learning: Training LLM Reasoners at the Edge of Capability

Reinforcement Learning-based post-training of Large Language Models (LLM) has been successfully applied to improve their reasoning capabilities. Existing pipelines primarily finetune LLMs on a fixed pool of problems specified prior to training using the GRPO loss. This is fundamentally limiting, as learning signal arises only when policy rollouts mix successes and failures, causing the useful portion of any fixed pool to quickly become stale as the model improves. To address this, we propose frontier learning, an open-ended post-training approach in which procedural generators are used online to continually produce informative training problems. It treats the generator's task-specific parameters as a search space and uses a regret signal to prioritize and explore frontier difficulty levels in order to focus training at the edge of the model's evolving reasoning capabilities. Across several reasoning tasks and model families, our approach consistently achieves higher relative gains over fixed-pool baselines, demonstrating that effective post-training requires not only selecting useful problems, but continually generating them at the edge of capability.

Robin Faro, Shyam Sundhar Ramesh, Ilija Bogunovic +1

University College London London, UK · University of Basel Basel, Switzerland

RL for Language Model Reasoning LLM Post-Training

Sep 27, 2026cs.CL

LLMs learn different forms of metacognition when trained to predict their own accuracy

Large language models are trained to always produce an answer, regardless of whether they possess the relevant knowledge, which leads them to fabricate facts. Prior work has shown that LLMs' confidence estimates correspond poorly to their actual performance, and that fine-tuning can substantially improve them. However, what models actually learn during such training remains poorly understood. We investigate how LLMs acquire metacognitive monitoring, the ability to know what one knows, by training 10 open-weight LLMs to predict their own accuracy on factual multiple-choice questions before answering them. We find that trained confidence reflects two distinct signals. While on questions close to the training data, it tracks the model's true accuracy, in other domains, it instead tracks output consistency: the concentration of the model's answer distribution. Output consistency tracking emerges early in training and generalizes across datasets, whereas accuracy tracking develops later and remains local to the training distribution. These results suggest that calibration training may not teach models to generally detect errors they commit confidently, and they raise broader questions about the nature of metacognition in artificial systems.

Nicolas Yax, Stefano Palminteri, Pierre-Yves Oudeyer

LNC2, INSERM, Paris, France · DEC, ENS, PSL, Paris, France · Flowers AI & CogSci Lab +1

Confidence Estimation in Language Models Language Model Calibration LLM Uncertainty Estimation LLM Post-Training Metacognition in Language Models

Sep 27, 2026cs.AI

How code helps different tasks? A decompositional lens on LLM post-training

Evaluating code data as a single corpus can obscure which types of code data benefit which models and downstream tasks. Effective data selection requires understanding both the benefits of individual categories and whether these benefits persist when categories are combined. We introduce a decompositional lens for studying these effects in LLM post-training. We first decompose an execution-verified code corpus into interpretable categories based on the computational patterns of its solutions. Through controlled fine-tuning experiments, we compare individual categories with a balanced mixture across instruction-tuned models on question answering, mathematics, and code generation. The resulting response maps reveal recurring gains in average question-answering performance, while the same category can improve one model or task and degrade another. The best-performing category also varies with the starting model and target task. We then compose compact mixtures guided by these results and examine whether benefits observed in individual categories persist under joint training. On selected model--task pairs, mixtures whose constituents each improve the target task outperform both their best constituent and full-corpus training while using roughly 10--15% of the full corpus. These exploratory findings illustrate a \emph{less is more} pattern and highlight how the value of code data in post training depends on which categories are combined for which model and task.

Zheng Yu, Yiwei Li, Yishen Chen +4

Sun Yat-sen University · Shenzhen Loop Area Institute · The Chinese University of Hong Kong, Shenzhen +1

LLM Fine-Tuning Instruction-Tuning Data Selection Training Data Selection LLM Post-Training LLM Code Generation LLM Reasoning

Sep 27, 2026cs.LG

A Cheap Verifier is Good Enough: LLM Post-training is Robust to Erroneous Rewards

When post-training large language models on tasks with semi-verifiable rewards, there are many factors (training steps, base model size, training order, data quality, verifier accuracy, etc.) that practitioners must contend with to maximize model performance. Yet, it remains unclear how well verifier agreement predicts post-training performance on such tasks. In this paper, we explore this question with over 11k H100 GPU-hours, across HealthBench and PRBench tasks in medical, legal, and finance domains. Across the tested domains, Qwen3 trainees (1.7B-8B on HealthBench; 8B on PRBench), evaluation splits, and frontier LLM reference judges (which we call golden verifiers), higher verifier agreement does not consistently identify the best training verifier. Expensive verifiers need not outperform inexpensive ones, and open-weight Gemma verifiers produce strong training outcomes. We compare two low-cost choices retrospectively -- a cost-reducing choice and a balanced choice -- with estimated grading cost reductions of 98.8%-99.7% relative to the golden grading protocols and average post-training score gaps of 1-3 points from the best evaluated training verifier. These averages include larger losses in individual settings; they do not establish that verifier choices are interchangeable.

Andreas Plesner, Curtis Northcutt, Francisco Guzmán +1

ETH Zurich · Handshake AI

Reward Modeling LLM Reliability Reinforcement Learning with Verifiable Rewards LLM Post-Training

Sep 24, 2026cs.AI

Search-Aware Reinforcement Learning for Multi-Component Query Understanding in Roblox Game Search

Query understanding (QU) plays a critical role in production search systems, translating raw user queries into search execution plans that drive downstream retrieval and ranking. While large language models (LLMs) have enabled QU to be framed as a structured multi-task generation problem (e.g., intent classification, query expansion), optimizing such models to produce search-engine-coupled outputs remains challenging: static, label-based supervision fails to capture how each component actually interacts with the underlying search pipeline to affect downstream performance. We present a search-aware reinforcement learning (RL) framework for QU based on a distill-then-RL paradigm. Teacher-student supervised fine-tuning (SFT) first yields a well-formed, schema-compliant policy initialization. The RL stage then optimizes each QU component with rewards derived from live interaction with the search engine, tailored to that component's operational role, rather than a single reward tied to the final search outcome. Experiments on Roblox search show that this component-specific optimization improves both per-component utility and downstream search quality, raising NDCG@20 by 8.9 points over the SFT policy and by 3.5 points over training with a single end-to-end reward.

Nayoung Choi, Shengjian Chen, Xiaokai Wei +6

Emory University Atlanta, Georgia, USA · Roblox Corporation San Mateo, California, USA

Reinforcement Learning RL for IR Information Retrieval LLM Post-Training

Sep 24, 2026cs.CL

Rufus-Air: An Open LLM Post-Training Recipe

Rufus-Air is an open and reproducible post-training recipe on GLM-4.5-Air-Base (106B-A12B), organized as a serial pipeline of eight stages: SFT, Reasoning RL, Coding RL, Instruction-Following RL, General Agent, Coding Agent, Search Agent, and RLHF. We document the data, reward design, infrastructure, stage order, and stagewise results needed to reproduce the recipe. Stages progress from basic to advanced capabilities and from hard, verifiable rewards to softer judge-based signals. Training builds on open-source components and public data, much of it used as released, without new human annotation or an in-house distillation teacher. Our main findings are that (i) diverse, high-quality SFT establishes a strong capability floor; (ii) difficulty filtering keeps RL prompts within a productive learning range; (iii) reward reliability provides a practical principle for ordering stages; and (iv) infrastructure and engineering choices are part of the recipe, not just an implementation detail. Rufus-Air improves over the official GLM-4.5-Air post-trained release and is competitive with similarly sized open models.

Chia-Yuan Chang, Renyuan Cheng, Rui Feng +19

Amazon

LLM Fine-Tuning RL for Language Model Reasoning RL Post-Training LLM Post-Training RL for LLM Agents

Sep 23, 2026cs.LG

Thinking Leakage: A Causal Audit of NoThink Post-Training in Hybrid Reasoning Models

Post-training hybrid reasoning models in NoThink mode has attracted growing interest as a way to improve performance while keeping inference fast. However, these gains may draw on thinking behavior already accessible through the base model's Think mode. We formulate this thinking leakage in a causal mediation framework and audit its contribution using bidirectional interventions along a simple base-derived activation direction. Across three models and three post-training methods on competition math benchmarks, we find that leakage is real, causal, and substantial: behavioral and representational analyses reveal shifts toward Think, steering the base model along this direction reproduces most of the post-training accuracy gain, and counter-steering a checkpoint removes a substantial share of what it gains. Across nine aligned checkpoints with positive NoThink gains, the resulting leakage ratio ranges from 42% to 79%. These interventions support a substantial causal contribution of thinking leakage. Our findings show that a post-training method's apparent advantage can therefore reflect greater drift toward Think, obscuring whether it improves capability within NoThink or more effectively re-invokes existing Think behavior.

Zehao Liu, Vasant G. Honavar

College of Information Sciences and Technology Pennsylvania State University

LLM Post-Training Causal Interventions in Language Models Large Reasoning Models

Sep 22, 2026cs.AI

Toolcompass: Guiding Tool Trialing, Not Suppressing It

Large language model (LLM) agents must generalize from tools seen during training to unseen tools at deployment. A key challenge is tool trialing, i.e., excessive trials waste the interaction budget, whereas selective trials enable exploration of unfamiliar tools. Existing outcome-based post-training leaves wasteful trials unguided, while turn-level supervision may suppress necessary exploration. We introduce ToolCompass, a post-training framework that guides tool trialing by organizing tool-call representations according to shared functions. Specifically, ToolCompass models each function class as a von Mises--Fisher distribution and jointly reduces intra-function variation across domains and increases inter-function separation. This structure transfers experience from seen tools to functionally similar unseen tools, directing exploration away from unrelated alternatives. ToolCompass requires no ground-truth call traces or unseen-tool access and incurs no inference overhead. Experiments on AppWorld and FTRL show consistent gains across GRPO, RFT, and DMPO. improves AppWorld OOD task success by up to 10.71 percentage points over vanilla post-training and performs best among competitive baselines on both benchmarks.

Junlin Fang, Chong Zhang, Do Nguyen-Thanh +3

College of Computing and Data Science, Nanyang Technological University · Zhejiang University · Australian Artificial Intelligence Institute, University of Technology Sydney +1

Representation Learning Tool-Augmented Language Model Agents LLM Tool Use LLM Post-Training Tool Use

Sep 21, 2026cs.CL

TelecomGPT-R1: Unified Post-Training for Reasoning Across Heterogeneous Telecom Tasks

Large language models (LLMs) offer great potential to automate a broad range of telecom engineering tasks by reasoning over standards, network configurations, mathematical models, source code, and operational logs. However, existing telecom LLMs struggle to reliably reason across these diverse tasks and data types. General-purpose LLMs often lack reliable grounding in telecom-specific knowledge, while telecom-specialized models are typically developed for narrower task families and exhibit limited multi-task performance. To fill this gap, we introduce TelecomGPT-R1, a family of open source unified telecom reasoning models structured around four complementary axes: protocol, knowledge, modeling, and fault. We first develop an axis-aware data generation framework that refines coarse public telecom artifacts into verified question-answer pairs and high quality chain-of-thought (CoT) reasoning trajectories, yielding a training corpus containing 104,880 examples. Building on this corpus, supervised fine-tuning (SFT) instills telecom knowledge and evidence-grounded reasoning patterns to overcome the cold start barrier for reinforcement learning (RL). We then apply dynamic sampling policy optimization (DAPO) with task-routed rubric rewards to keep RL updates informative and stable across heterogeneous telecom reasoning tasks. These rewards decompose axis-specific CoT traces into verifiable reasoning units and combine grounded dense process credit with outcome correctness, allowing RL to learn generalizable problem solving behaviors from verifiable telecom evidence. We release the TelecomGPT-R1 models and a reproducible training recipe to support further community development. Evaluations on seven benchmarks of the GSMA Open Telco Leaderboard show that the open-source TelecomGPT-R1-27B achieves an 89.64% mean score, outperforming leading proprietary models, including GPT-5, Claude, and Gemini.

Bohao Wang, Chenwei Wu, Hang Zou +7

College of Information Science and Electronic Engineering, Zhejiang University, 310027 Hangzhou, China · Research Institute for Digital Future, Khalifa University, P O Box 127788, Abu Dhabi, UAE · Department of Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI 48109-2122, USA +1

LLM Grounding RL for Language Model Reasoning LLM Post-Training

Sep 21, 2026cs.LG

SupportCal: Label-Free Calibration of Post-Trained LLMs via Reference Support and Corroboration

Post-training often improves task performance but can degrade confidence calibration, leaving post-trained language models (PoLMs) more overconfident than their corresponding pretrained language models (PLMs). Because task-specific labeled calibration data can be costly or unavailable, the corresponding PLM provides a natural label-free reference for post-hoc calibration. Prior agreement-gated PLM-referenced calibration fits a scalar temperature using only examples on which the PoLM and its PLM reference agree, excluding disagreement examples because direct alignment can drive the fitted temperature excessively high and induce under-confidence. We revisit this binary treatment. A controlled reintroduction diagnostic reveals a non monotonic aggregate effect: admitting a moderate fraction of disagreement examples can improve calibration, whereas the benefit diminishes as unit weight inclusion approaches the full disagreement set. We introduce SupportCal, a label-free post-hoc method that retains agreement examples at unit weight and assigns disagreement examples continuous weights based on the own-base PLM's relative support and corroboration from pretrained references selected from a size-compatible candidate pool. We further characterize when the resulting weighted objective admits a finite optimal temperature. Across MedMCQA and MathQA, SupportCal yields lower mean ECE than the agreement-only baseline for nearly all evaluated target-model configurations; supplementary TweetEval Sentiment results show the same pattern on a fixed-label classification task.

Linhan Luo, Lequan Lin, Dai Shi +3

The University of Sydney · University of Cambridge · The University of Adelaide

Post-Hoc Calibration Language Model Calibration LLM Post-Training

Sep 20, 2026cs.CL

Measuring the Assistant's Harmlessness Preferences on the User Turn

Post-training turns a general next-token predictor into a chat model with a persistent assistant persona. If that persona is a character the model plays only on its own turns, its preferences should govern what the assistant says, not what the model predicts other speakers will say. We test this boundary and find that it does not hold: a safety-relevant preference of the assistant---for harmless over harmful tasks---shapes the model's predictions even on the user's turn, where the assistant is not the one speaking. We find that this preference is small or near-zero in pretrained base models, that it emerges through post-training, replicated across open-weight model families, grows with scale, and can be moved by narrow finetuning that never touches user turns. We claim that this is evidence that post-training does not merely install a shallow assistant persona, but instead generalises beyond just the local assistant turn, into the model's representation of the user.

Jord Nguyen

Language Model Safety Evaluation LLM Safety LLM Post-Training

Sep 15, 2026cs.CL

Style-Debiased DPO: Updating LLM Knowledge with Factuality-Aware Synthetic Preference Data

Continued pretraining (CPT) with data augmentation such as paraphrasing can store inside a large language model (LLM) the knowledge of a small source corpus. The stored knowledge, however, is not always retrieved correctly. We study the eliciting side rather than the storing side: we use preference optimization, which learns from pairs of a preferred (chosen) and a dispreferred (rejected) response, so that the model elicits its stored knowledge more accurately. One proposed approach takes the model's own erroneous response as rejected and the gold answer as chosen, so as to suppress the error. When the target knowledge is partially known, however, most of these rejected responses are factually correct. Using direct preference optimization (DPO) then pushes down rejected responses that contain correct knowledge and differ from the chosen answer only in style, such as length and wording. We propose style-debiased DPO (SD-DPO), which scores whether the rejected response of each pair is factually correct, inverts the preference of such pairs, and weights them so that the learning signal due to differences in style cancels out as a whole. We first test whether, on top of EntiGraph, a representative storing-side method that runs CPT on text synthesized from the corpus, our method adds accuracy efficiently. On QuALITY, the reading-comprehension QA benchmark on which EntiGraph was evaluated, SD-DPO exceeds a baseline we CPT on EntiGraph's synthetic data from the same base model and evaluate with the same procedure. The training tokens this requires are a few dozen times fewer than the additional CPT needed for the same gain. For knowledge updating, the main goal of this work, we use AToKE, a knowledge-editing benchmark for facts that change over time. There, SD-DPO reaches an overall accuracy of 0.982 and answers with the new or the old fact according to the queried period.

Takayuki Yamamoto, Daisuke Kawahara

Waseda University

Knowledge Editing Direct Preference Optimization Factual Knowledge in Language Models LLM Post-Training

Sep 14, 2026cs.CL

Inoculation Midtraining with Learned Neologisms

Large language models (LLMs) often learn both desirable and undesirable properties during post-training. We study whether midtraining, an earlier training stage, can shape which of these properties later generalise. We introduce Inoculation Midtraining, a technique that teaches a base model that unsafe behaviour belongs to a designated <quarantine_token> context, as indicated by the <quarantine_token> neologism (a new token) introduced during midtraining, and then post-trains the model on unsafe data within that context. We then evaluate the model outside the context, with the <quarantine_token> neologism excluded from the system prompt. Across supervised fine-tuning and reinforcement learning post-training regimes, we find that Inoculation Midtraining can reduce misalignment while preserving the transfer of benign data properties (e.g., speaking in German or Shakespearean prose). However, our approach does not outperform standard Inoculation Prompting, is sensitive to training configuration, and produces a leaky boundary that nearby contextual cues can reactivate. These results show that inoculation with a learned association introduced via midtraining can shape selective generalisation. Still, more work is needed before this approach can become a load-bearing component in a developer's safety framework.

Kyle O'Brien, Edward James Young, Puria Radmard +4

Geodesic Research · OpenAI · UK AI Security Institute

LLM Alignment LLM Safety LLM Post-Training

Sep 14, 2026cs.AI

Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work

Co-work agents execute complex workflows that combine information gathering, tool use, coding, and file manipulation across many model invocations. Because cost and latency accumulate over the full episode, their practical value depends not only on peak capability but also on how efficiently that capability is delivered. Yet many steps in everyday work emphasize state tracking, coordination, recovery, and follow-through rather than frontier-scale reasoning. We present Occamy-1.0, a cost-efficient co-work model obtained by further training the post-trained Qwen3.6-35B-A3B checkpoint. We construct execution-grounded data and environments, capture replayable long-horizon trajectories across multiple harnesses, and use staged post-training to develop and consolidate complementary execution capabilities. Across a broad suite of co-work benchmarks, Occamy-1.0 is consistently among the strongest comparably sized models and remains competitive with substantially larger frontier systems on several tasks. Under our stated evaluation and pricing protocol, its aggregate performance across four representative benchmarks places it at the low-cost knee of the observed cost--performance Pareto frontier. Supporting evaluations in tool calling, coding, and instruction following further show that this specialization preserves broad agentic capability. We release the model weights and a subset of the training data to support research on practical co-work agents and agentic post-training.

Wenhui Chen, Shiwen Cheng, Hao Dong +40

LLM Agent Training LLM Post-Training Long-Horizon LLM Agents