We present **LLaDA-o**, an effective and length-adaptive omni diffusion model for multimodal understanding and generation. LLaDA-o is built on a Mixture of Diffusion (MoD) framework that decouples discrete masked diffusion for text understanding and continuous diffusion for visual generation, while coupling them through a shared, simple, and efficient attention backbone that reduces redundant computation for fixed conditions. Building on MoD, we further introduce a data-centric length adaptation strategy that enables flexible-length decoding in multimodal settings without architectural changes. Extensive experiments show that LLaDA-o achieves strong performance on multimodal understanding and generation benchmarks, reaching 87.04 on DPG-Bench and 0.86 on GenEval. To support future work on omni-diffusion models, we release the model and training/evaluation code at https://github.com/ML-GSAI/LLaDA-o.

Extending large language models (LLMs) to low-resource languages often incurs an “align- ment tax”: improvements in the target lan- guage come at the cost of catastrophic forget- ting in general capabilities. We argue that this trade-off arises from the rigidity of supervised fine-tuning (SFT), which enforces token-level surface imitation on narrow and biased data distributions. To address this limitation, we propose a semantic-space alignment paradigm powered by Group Relative Policy Optimiza- tion (GRPO), where the model is optimized us- ing embedding-level semantic rewards rather than likelihood maximization. This objective encourages meaning preservation through flex- ible realizations, enabling controlled updates that reduce destructive interference with pre- trained knowledge. We evaluate our approach on Tibetan–Chinese machine translation and Ti- betan headline generation. Experiments show that our method acquires low-resource capa- bilities while markedly mitigating alignment tax, preserving general competence more effec- tively than SFT. Despite producing less rigid surface overlap, semantic RL yields higher se- mantic quality and preference in open-ended generation, and few-shot transfer results indi- cate that it learns more transferable and ro- bust representations under limited supervision. Overall, our study demonstrates that reinforce- ment learning with semantic rewards provides a safer and more reliable pathway for inclusive low-resource language expansion.

Despite the remarkable success of Large Language Models (LLMs) in Machine Translation (MT), the scarcity of high-quality parallel corpora and the prohibitive cost of their acquisition constrain scalability. To this end, we propose Learning to Translate by Translating (LTT), an LLM-driven dual-learning framework that enables autonomous translation, achieving an 80.42% performance improvement over the base model. By adapting the cycle-consistency principle to the generative paradigm, LTT eliminates the need for parallel data. It employs a robust semantic-aware reward function that balances adequacy with reconstruction fidelity, effectively mitigating the reward hacking issues inherent in traditional unsupervised MT. Relying solely on monolingual data, our 8B model consistently outperforms significantly larger models (70B+) in low-resource settings and achieves parity with state-of-the-art supervised baselines on mainstream benchmarks. LTT thus offers a scalable, data-efficient paradigm for autonomous machine translation.

Low-resource target-language generation is often limited by scarce parallel data, while high-resource source-language monolingual data is abundant but difficult to use with standard supervised fine-tuning. We propose Source-Grounded Semantic Reinforcement Learning (SG-SRL), a resource-utilization framework that converts source-language monolingual data into cross-lingual semantic supervision. SG-SRL performs reference-free reinforcement learning (RL) on source-language data with a cross-lingual semantic reward, followed by lightweight supervised recovery. Although direct reward optimization induces verbosity-based reward hacking, recovery restores fluency, conciseness, and task format without erasing the learned semantic alignment. On Chinese-to-Thai generation, SG-SRL outperforms cold-start SFT, source-only continued pretraining, and synthetic-data SFT; human preference and training-independent metrics corroborate the gains. Recovery ablations further show that semantic retention and form restoration are jointly controlled by recovery data and intensity. An in-domain Chinese-to-Tibetan experiment shows that the reward interface can use a contrastively trained encoder when a capable multilingual reranker is unavailable, while also delimiting the framework’s dependence on reward-model discrimination.

Large Language Model (LLM) evaluation is frequently challenged by prompt sensitivity, which limits the reliability of benchmarks relying on a single template. To address this, we systematically investigate a multi-prompt evaluation framework that aggregates performance across diverse templates, leveraging a Quality-Diversity (QD) prompt selection mechanism to ensure both variety and fidelity. Through extensive empirical analysis, we uncover a novel Evaluation Scaling Law: evaluation reliability scales consistently with template diversity. Specifically, our experiments demonstrate that evaluation variance diminishes as the scale of prompts increases, while ranking monotonicity exhibits a log-linear improvement relative to the number of templates. By decoupling a model’s intrinsic capability from template-specific fluctuations, our framework facilitates more reliable assessments of LLMs, thereby providing a solid foundation for benchmarking that enables more effective monitoring of training dynamics and rigorous comparisons among state-of-the-art models.

Masked diffusion language models present a promising paradigm for language modeling, yet the systematic theoretical analysis and comprehensive empirical validation of their alignment on general tasks remain relatively underexplored. In this paper, we identify the primary challenge for this problem: the high variance in Evidence Lower Bound (ELBO)-based likelihood estimates required for preference optimization. To address this issue, we propose Variance-Reduced Preference Optimization (VRPO), a framework that formally analyzes the bias and variance of the preference optimization loss and gradient based on Direct Preference Optimization, showing both are governed by a score-estimator variance. Building on this foundation, we introduce multiple unbiased variance reduction strategies, including optimal budget allocation and antithetic sampling, to improve alignment performance. We demonstrate the effectiveness of VRPO by applying it to LLaDA, a large diffusion language model. The resulting model, LLaDA 1.5, consistently outperforms its SFT-only predecessor consistently across various general benchmarks, such as mathematics (GSM8K +4.7), coding (HumanEval +3.0, MBPP +1.8), and alignment (IFEval +4.0, Arena-Hard +4.3). Furthermore, LLaDA 1.5 demonstrates a highly competitive mathematical performance compared to other strong language MDMs and ARMs. Our model is available at https://huggingface.co/GSAI-ML/LLaDA-1.5.

Users increasingly rely on Large Language Models (LLMs) for Deep Research, using them to synthesize diverse sources into structured reports that support understanding and action. In this context, the practical reliability of such reports hinges on logical quality: whether the report’s claims and arguments are explicitly supported and can be trusted as a basis for downstream use, rather than merely appearing fluent or informative. However, current evaluation frameworks largely overlook this requirement. To bridge this gap, we introduce ReportLogic, a benchmark that quantifies report-level logical quality through a reader-centric lens of auditability. Specifically, ReportLogic adopts a hierarchical taxonomy that evaluates whether readers can (1) trace an on-topic report structure with a unified analytical arc (Macro-Logic), (2) understand the progression with necessary context (Expositional-Logic), and (3) verify conclusions via explicit claim–support (Structural-Logic). Based on this taxonomy, we construct a human-annotated rubric-guided dataset and train an open-source LogicJudge for scalable evaluation. We further evaluate judge robustness via adversarial attacks, showing that off-the-shelf LLM judges are frequently influenced by superficial cues (e.g., verbosity), and reasoning modes can mask broken support relations. Overall, our results provide actionable guidance for building more robust logic evaluators and improving the logical reliability of LLM-generated reports.

This paper poses two critical issues in evaluating base models (without post-training): (1) Unstable evaluation during training: in the early stages of pre-training, the models lack the capability to answer questions as required, leading to unstable evaluation results. This instability makes it difficult to provide solid conclusions to guide the training, especially for key experiments such as data ablation and scaling law. (2) Inconsistency between base and instruct models: base models generally exhibit poorer evaluation performance compared to corresponding instruct models. This gap poses a challenge for assessing whether a base model with better evaluation can truly lead to a better instruct model. To address these issues, we propose Base model Oriented Systematic Evaluation (BOSE), a method specifically designed to optimize the evaluation of base models. Specifically, BOSE introduces two key innovations: In-Context Light-instruction Prompt (ICLiP) for open-ended tasks and Blank-ppl for multi-choice tasks with candidate options, which transforms the standard perplexity (ppl) metric into a fill-in-the-blank format to mitigate early-stage evaluation fluctuations. Furthermore, we are the first to propose Kendall’s rank correlation to quantitatively measure the evaluation stability and consistency. Experimental results demonstrate that BOSE significantly enhances both the stability of evaluations during pre-training and the consistency between base and instruct models, thereby providing more reliable guidance for the LLMs’ training.

Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, yet they often refuse to answer legitimate queries—a phenomenon known as overrefusal. Overrefusal typically stems from over-conservative safety alignment, causing models to treat many reasonable prompts as potentially risky. To systematically understand this issue, we probe and leverage the models’ safety decision boundaries to analyze and mitigate overrefusal. Our findings reveal that overrefusal is closely tied to misalignment at these boundary regions, where models struggle to distinguish subtle differences between benign and harmful content. Building on these insights, we present RASS, an automated framework for prompt generation and selection that strategically targets overrefusal prompts near the safety boundary. By harnessing steering vectors in the representation space, RASS efficiently identifies and curates boundary-aligned prompts, enabling more effective and targeted mitigation of overrefusal. This approach not only provides a more precise and interpretable view of model safety decisions but also seamlessly extends to multilingual scenarios. We have explored the safety decision boundaries of various LLMs and construct the MORBench evaluation set to facilitate robust assessment of model safety and helpfulness across multiple languages. Code and datasets are available at https://github.com/Master-PLC/RASS.