What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents

Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignment remains challenging: a single success-or-failure signal must be distributed across many actions. Existing methods rely on trajectory-level rewards or proxy signals, without fully leveraging per-step environmental feedback. Multi-turn agent settings are underexplored, where feedback can include error messages, page changes, observations, or reference trajectories. We systematically study five feedback sources and two insertion granularities and introduce SERL, a selective environment-reweighted learning framework. SERL uses the task reward to determine update direction, while environment feedback adjusts placement and magnitude, focusing on critical actions. On ALFWorld and WebShop, SERL achieves 90.0% and 80.1% success, outperforming strong RL and distillation baselines. Analysis shows that grounded, action-relevant feedback at meaningful points consistently outperforms indiscriminate use of longer or richer context.

Beyond Mode Collapse: Distribution Matching for Diverse Reasoning

On-policy reinforcement learning methods like GRPO suffer from mode collapse: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (Distribution-Matching Policy Optimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO constructs a group level target distribution over sampled trajectories proportional to their rewards, then aligns the policy distribution to this target. This provides mode-covering behavior without requiring sampling from the intractable global target distribution, enabling sustained exploration throughout training. We validate DMPO on NP-hard combinatorial optimization, where exponentially many feasible solutions exist but only a few approach optimality, an ideal testbed for evaluating exploration. DMPO achieves 43.9% Quality Ratio on text-based NP-Bench (vs. GRPO's 40.1%) and 43.1% on vision-based NP-Bench (vs. 38.4%), demonstrating 9% and 12% relative improvements respectively. These gains generalize to mathematical reasoning (+2.0%) and out-of-domain tasks (+2.3%), showing that diversity-preserving training enhances general reasoning capabilities across modalities. Our work establishes distribution matching as a practical, principled approach to preventing mode collapse in on-policy RL, with consistent quality improvements demonstrating sustained exploration across diverse reasoning tasks.

Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data

Large language models (LLMs) rely on web-scale corpora for pre-training. The noise inherent in these datasets tends to obscure meaningful patterns and ultimately degrade model performance. Data curation mitigates but cannot eliminate such noise, so pre-training corpora remain noisy in practice. We therefore study whether a lightweight pre-pre-training (PPT) stage based on synthetic data with learnable temporal structure helps resist noisy data during the pre-training (PT) stage. Across various corruption settings, our method consistently improves robustness to noise during PT, with larger relative gains at higher noise levels. For a 1B-parameter model, a synthetic PPT stage with only 65M tokens achieves the same final loss as the baseline while using up to 49\% fewer natural-text PT tokens across different noise levels. Mechanistic analyses suggest PPT does not immediately suppress attention to noisy tokens. Rather, PPT-initialized models gradually downweight attention between corrupted tokens during noisy PT. This indicates that synthetic PPT inhibits noise self-modeling and shapes the subsequent optimization trajectory. Code is available at https://github.com/guox18/formal-language-prepretraining.

Kernel-Smith: A Unified Recipe for Evolutionary Kernel Optimization

We present Kernel-Smith, a framework for high-performance GPU kernel and operator generation that combines a stable evaluation-driven evolutionary agent with an evolution-oriented post-training recipe. On the agent side, Kernel-Smith maintains a population of executable candidates and iteratively improves them using an archive of top-performing and diverse programs together with structured execution feedback on compilation, correctness, and speedup. To make this search reliable, we build backend-specific evaluation services for Triton on NVIDIA GPUs and Maca on MetaX GPUs. On the training side, we convert long-horizon evolution trajectories into step-centric supervision and reinforcement learning signals by retaining correctness-preserving, high-gain revisions, so that the model is optimized as a strong local improver inside the evolutionary loop rather than as a one-shot generator. Under a unified evolutionary protocol, Kernel-Smith-235B-RL achieves state-of-the-art overall performance on KernelBench with Nvidia Triton backend, attaining the best average speedup ratio and outperforming frontier proprietary models including Gemini-3.0-pro and Claude-4.6-opus. We further validate the framework on the MetaX MACA backend, where our Kernel-Smith-MACA-30B surpasses large-scale counterparts such as DeepSeek-V3.2-think and Qwen3-235B-2507-think, highlighting potential for seamless adaptation across heterogeneous platforms. Beyond benchmark results, the same workflow produces upstream contributions to production systems including SGLang and LMDeploy, demonstrating that LLM-driven kernel optimization can transfer from controlled evaluation to practical deployment.

daVinci-LLM:Towards the Science of Pretraining

The foundational pretraining phase determines a model's capability ceiling, as post-training struggles to overcome capability foundations established during pretraining, yet it remains critically under-explored. This stems from a structural paradox: organizations with computational resources operate under commercial pressures that inhibit transparent disclosure, while academic institutions possess research freedom but lack pretraining-scale computational resources. daVinci-LLM occupies this unexplored intersection, combining industrial-scale resources with full research freedom to advance the science of pretraining. We adopt a fully-open paradigm that treats openness as scientific methodology, releasing complete data processing pipelines, full training processes, and systematic exploration results. Recognizing that the field lacks systematic methodology for data processing, we employ the Data Darwinism framework, a principled L0-L9 taxonomy from filtering to synthesis. We train a 3B-parameter model from random initialization across 8T tokens using a two-stage adaptive curriculum that progressively shifts from foundational capabilities to reasoning-intensive enhancement. Through 200+ controlled ablations, we establish that: processing depth systematically enhances capabilities, establishing it as a critical dimension alongside volume scaling; different domains exhibit distinct saturation dynamics, necessitating adaptive strategies from proportion adjustments to format shifts; compositional balance enables targeted intensification while preventing performance collapse; how evaluation protocol choices shape our understanding of pretraining progress. By releasing the complete exploration process, we enable the community to build upon our findings and systematic methodologies to form accumulative scientific knowledge in pretraining.

Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale

We introduce Intern-S1-Pro, the first one-trillion-parameter scientific multimodal foundation model. Scaling to this unprecedented size, the model delivers a comprehensive enhancement across both general and scientific domains. Beyond stronger reasoning and image-text understanding capabilities, its intelligence is augmented with advanced agent capabilities. Simultaneously, its scientific expertise has been vastly expanded to master over 100 specialized tasks across critical science fields, including chemistry, materials, life sciences, and earth sciences. Achieving this massive scale is made possible by the robust infrastructure support of XTuner and LMDeploy, which facilitates highly efficient Reinforcement Learning (RL) training at the 1-trillion parameter level while ensuring strict precision consistency between training and inference. By seamlessly integrating these advancements, Intern-S1-Pro further fortifies the fusion of general and specialized intelligence, working as a Specializable Generalist, demonstrating its position in the top tier of open-source models for general capabilities, while outperforming proprietary models in the depth of specialized scientific tasks.

AI Can Learn Scientific Taste

Great scientists have strong judgement and foresight, closely tied to what we call scientific taste. Here, we use the term to refer to the capacity to judge and propose research ideas with high potential impact. However, most relative research focuses on improving an AI scientist's executive capability, while enhancing an AI's scientific taste remains underexplored. In this work, we propose Reinforcement Learning from Community Feedback (RLCF), a training paradigm that uses large-scale community signals as supervision, and formulate scientific taste learning as a preference modeling and alignment problem. For preference modeling, we train Scientific Judge on 700K field- and time-matched pairs of high- vs. low-citation papers to judge ideas. For preference alignment, using Scientific Judge as a reward model, we train a policy model, Scientific Thinker, to propose research ideas with high potential impact. Experiments show Scientific Judge outperforms SOTA LLMs (e.g., GPT-5.2, Gemini 3 Pro) and generalizes to future-year test, unseen fields, and peer-review preference. Furthermore, Scientific Thinker proposes research ideas with higher potential impact than baselines. Our findings show that AI can learn scientific taste, marking a key step toward reaching human-level AI scientists.

Rethinking Multiple-Choice Questions for RLVR: Unlocking Potential via Distractor Design

Reinforcement Learning with Verifiable Rewards (RLVR) significantly enhances the reasoning capabilities of Large Language Models. When applied to RLVR, Multiple-Choice Questions (MCQs) offer a scalable source of verifiable data but risk inducing reward hacking, where models shortcut reasoning via random guessing or simple elimination. Current approaches often mitigate this by converting MCQs to open-ended formats, thereby discarding the contrastive signal provided by expert-designed distractors. In this work, we systematically investigate the impact of option design on RLVR. Our analysis highlights two primary insights: (1) Mismatches in option counts between training and testing degrade performance. (2) Strong distractors effectively mitigate random guessing, enabling effective RLVR training even with 2-way questions. Motivated by these findings, we propose Iterative Distractor Curation (IDC), a framework that actively constructs high-quality distractors to block elimination shortcuts and promote deep reasoning. Experiments on various benchmarks demonstrate that our method effectively enhances distractor quality and yields significant gains in RLVR training compared to the original data.

InternVL-U: Democratizing Unified Multimodal Models for Understanding, Reasoning, Generation and Editing

Unified multimodal models (UMMs) that integrate understanding, reasoning, generation, and editing face inherent trade-offs between maintaining strong semantic comprehension and acquiring powerful generation capabilities. In this report, we present InternVL-U, a lightweight 4B-parameter UMM that democratizes these capabilities within a unified framework. Guided by the principles of unified contextual modeling and modality-specific modular design with decoupled visual representations, InternVL-U integrates a state-of-the-art Multimodal Large Language Model (MLLM) with a specialized MMDiT-based visual generation head. To further bridge the gap between aesthetic generation and high-level intelligence, we construct a comprehensive data synthesis pipeline targeting high-semantic-density tasks, such as text rendering and scientific reasoning, under a reasoning-centric paradigm that leverages Chain-of-Thought (CoT) to better align abstract user intent with fine-grained visual generation details. Extensive experiments demonstrate that InternVL-U achieves a superior performance - efficiency balance. Despite using only 4B parameters, it consistently outperforms unified baseline models with over 3x larger scales such as BAGEL (14B) on various generation and editing tasks, while retaining strong multimodal understanding and reasoning capabilities.

Mousse: Rectifying the Geometry of Muon with Curvature-Aware Preconditioning

Recent advances in spectral optimization, notably Muon, have demonstrated that constraining update steps to the Stiefel manifold can significantly accelerate training and improve generalization. However, Muon implicitly assumes an isotropic optimization landscape, enforcing a uniform spectral update norm across all eigen-directions. We argue that this "egalitarian" constraint is suboptimal for Deep Neural Networks, where the curvature spectrum is known to be highly heavy-tailed and ill-conditioned. In such landscapes, Muon risks amplifying instabilities in high-curvature directions while limiting necessary progress in flat directions. In this work, we propose \textbf{Mousse} (\textbf{M}uon \textbf{O}ptimization \textbf{U}tilizing \textbf{S}hampoo's \textbf{S}tructural \textbf{E}stimation), a novel optimizer that reconciles the structural stability of spectral methods with the geometric adaptivity of second-order preconditioning. Instead of applying Newton-Schulz orthogonalization directly to the momentum matrix, Mousse operates in a whitened coordinate system induced by Kronecker-factored statistics (derived from Shampoo). Mathematically, we formulate Mousse as the solution to a spectral steepest descent problem constrained by an anisotropic trust region, where the optimal update is derived via the polar decomposition of the whitened gradient. Empirical results across language models ranging from 160M to 800M parameters demonstrate that Mousse consistently outperforms Muon, achieving around $\sim$12\% reduction in training steps with negligible computational overhead.