Large language models (LLMs) face significant computational and memory challenges, making extremely low-bit quantization crucial for their efficient deployment. In this work, we introduce SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size, a novel framework that enables extremely low-bit quantization of LLMs while preserving their linguistic reasoning capabilities. A distinctive feature of SDQ-LLM is the continuous adjustability of the Over-Sampling Ratio (OSR), enabling dynamic adaptation to memory or VRAM constraints by selecting fractional OSR (e.g. 2.5 times) for an optimal trade-off between model size and accuracy. SDQ-LLM uses upsampling combined with Sigma-Delta Quantizer to binarize or ternarize LLMs weights, encoding high-precision parameters into 1-bit or 1.58-bit representations, replacing the multiplication operations within linear layers with addition. This approach significantly enhances inference efficiency under extremely low-bit quantization. To further reduce the loss of quantization precision, we incorporate Hadamard-based weight smoothing prior to quantization, improving the stability and robustness of the weight representations. Furthermore, to fully leverage the continuity of the OSR and reduce precision loss, recognizing the correlation between quantization sensitivity and weight variance, we propose a fine-grained, layer- and linear-wise OSR allocation strategy, MultiOSR. This strategy distributes OSR both across layers and within each layer, based on weight variance and parameter scale. Finally, extensive experiments on OPT and LLaMA model families demonstrate that SDQ-LLM achieves a more efficient and high-precision performance even under highly aggressive low-OSR settings. Our code is available at https://github.com/Dreamlittlecat/LLM-Quant-Factory.

Diffusion large language models (dLLMs), which offer bidirectional context and flexible masked-denoising generation, are emerging as a compelling alternative to autoregressive (AR) LLMs. However, like AR LLMs, their model sizes continue to grow, motivating weight compression for deployment. Although post-training quantization (PTQ) is effective for AR LLMs, directly transferring it to dLLMs at 2-bit leads to unsatisfactory performance. To tackle these challenges, we propose Quant-dLLM, an ultra-low-bit PTQ framework tailored to dLLMs. Since masked-denoising activations in dLLMs differ from the fully visible signals assumed by standard PTQ methods, we introduce Masked Calibration Simulation (MCS) to align calibration with the timestep-dependent masking, which yields more reliable calibrations. Moreover, we propose a Data-aware Any-order Quantizer (DAQ) that learns ultra-low-bit weight representations via an optimization algorithm. It performs iterative approximation guided by our simulated calibration data. In addition, under a strict 2-bit budget, we introduce Adaptive Blockwise Mixed Precision (ABMP), a sensitivity-based precision allocation scheme that adaptively assigns bit width across channel groups. When restricted to 2-bit precision, Quant-dLLM consistently achieves higher accuracy than state-of-the-art (SOTA) AR-transfer PTQ methods on dLLMs. The code and models will be available at: https://github.com/ZTA2785/Quant-dLLM.

Decision boundary, the subspace of inputs where a machine learning model assigns equal classification probabilities to two classes, is pivotal in revealing core model properties and interpreting behaviors. While analyzing the decision boundary of large language models (LLMs) has raised increasing attention recently, constructing it for mainstream LLMs remains computationally infeasible due to the enormous vocabulary-sequence sizes and the auto-regressive nature of LLMs. To address this issue, in this paper we propose Decision Potential Surface (DPS), a new notion for analyzing LLM decision boundary. DPS is defined on the confidences in distinguishing different sampling sequences for each input, which naturally captures the potential of decision boundary. We prove that the zero-height isohypse in DPS is equivalent to the decision boundary of an LLM, with enclosed regions representing decision regions. By leveraging DPS, for the first time in the literature, we propose an approximate decision boundary construction algorithm, namely $K$-DPS, which only requires K-finite times of sequence sampling to approximate an LLM's decision boundary with negligible error. We theoretically derive the upper bounds for the absolute error, expected error, and the error concentration between K-DPS and the ideal DPS, demonstrating that such errors can be trade-off with sampling times. Our results are empirically validated by extensive experiments across various LLMs and corpora.

Diffusion language models promise bidirectional context and infilling capabilities that autoregressive coders lack, yet practical systems remain heavyweight. We introduce CoDA, a 1.7B-parameter diffusion coder trained on TPU with a fully open-source training pipeline. CoDA pairs large-scale diffusion pre-training with code-centric mid-training and instruction tuning, enabling confidence-guided sampling that keeps inference latency competitive. On Humaneval, MBPP, and EvalPlus, CoDA-1.7B-Instruct matches or surpasses diffusion models up to 7B parameters. Our release includes model checkpoints, evaluation harnesses, and TPU training pipelines to accelerate research on lightweight diffusion-based coding assistants.

The prevailing paradigm for enhancing the reasoning abilities of LLMs revolves around post-training on high-quality, reasoning-intensive data. While emerging literature suggests that reasoning data is increasingly incorporated also during the mid-training stage-a practice that is relatively more proprietary and less openly characterized-the role of such data in pretraining remains unclear. In particular, due to the opaqueness of pretraining corpora in most frontier models, the effect of reasoning data introduced at different phases of pre- and/or post-training is relatively less reported in the scientific literature. This raises several important questions: Is adding reasoning data earlier during pretraining any better than introducing it during post-training? Could earlier inclusion risk overfitting and harm generalization, or instead establish durable foundations that later fine-tuning cannot recover? We conduct the first systematic study of how reasoning data-varying in scale, diversity, and quality-affects LLM performance when introduced at different stages of training. We find that front-loading reasoning data into pretraining is critical (19% avg gain), establishing foundational capabilities that cannot be fully replicated by later-stage SFT, even with more data. We uncover an asymmetric principle for optimal data allocation: pretraining benefits most from broad diversity in reasoning patterns (11% avg gain), while SFT is more sensitive to data quality (15% avg gain). We show that high-quality pretraining data has latent effects, activated only after SFT, and that naively scaling SFT data can be detrimental, washing away the benefits of early reasoning injection. Our results challenge the conventional separation of language modeling and reasoning, providing a principled guide for strategically allocating data across the entire training pipeline to build more capable models.

Zero-Shot Learning is an important paradigm within General-Purpose Artificial Intelligence Systems, particularly in those that operate in open-world scenarios where systems must adapt to new tasks dynamically. Semantic spaces play a pivotal role as they bridge seen and unseen classes, but whether human-annotated or generated by a machine learning model, they often contain noisy, redundant, or irrelevant attributes that hinder performance. To address this, we introduce a partitioning scheme that simulates unseen conditions in an inductive setting (which is the most challenging), allowing attribute relevance to be assessed without access to semantic information from unseen classes. Within this framework, we study two complementary feature-selection strategies and assess their generalisation. The first adapts embedded feature selection to the particular demands of ZSL, turning model-driven rankings into meaningful semantic pruning; the second leverages evolutionary computation to directly explore the space of attribute subsets more broadly. Experiments on five benchmark datasets (A WA2, CUB, SUN, aPY, FLO) show that both methods consistently improve accuracy on unseen classes by reducing redundancy, but in complementary ways: RFS is efficient and competitive though dependent on critical hyperparameters, whereas GA is more costly yet explores the search space more broadly and avoids such dependence. These results confirm that semantic spaces are inherently redundant and highlight the proposed partitioning scheme as an effective tool to refine them under inductive conditions.

Recent studies on reasoning models explore the meta-awareness of language models, the ability to know how to think by itself. We argue that large reasoning models lack this meta-awareness property by proving severe misalignment between true rollouts and predicted meta information. We posit that aligning meta-prediction with true rollouts will lead to significant performance gains. To verify this hypothesis, we design a training pipeline that boosts Meta-Awareness via Self-Alignment (MASA), and prove that enhanced meta-awareness directly translates to improved accuracy. Unlike existing meta-cognitive reasoning models, our method does not require external training sources but leverages self-generated signals to train meta-awareness. Moreover, our method enables efficient training by i) filtering out zero-variance prompts that are either trivial or unsolvable and ii) cutting off lengthy rollouts when they are unlikely to lead to correct answers. The results are inspiring: our strategy yields significant improvements in both accuracy and training efficiency on in-domain tasks and shows strong generalization to out-of-domain benchmarks. More specifically, our method can speed up GRPO training by over 1.28x to reach the same performance, and achieve a 19.3% gain in accuracy on AIME25, and a 6.2 % average gain over six mathematics benchmarks. Training with meta-cognitive guidance enhances out-of-domain generalization, giving a 3.87 % boost on GPQA-Diamond and a 2.08 % overall accuracy gain across 13 benchmarks spanning logical, scientific, and coding domains.

Pruning is critical for scaling large language models (LLMs). Global pruning achieves strong performance but requires $\mathcal{O}(N)$ memory, which is infeasible for billion-parameter models. Local pruning reduces GPU memory usage to that of a single layer by pruning layers independently, but it neglects inter-layer dependencies and often leads to suboptimal performance in high-sparsity regimes. Unlike unstructured pruning, structured pruning produces regular sparsity patterns that align well with GPU kernels and library optimizations, making it more hardware-efficient. However, structured pruning typically relies on global pruning, since structured patterns are more prone to severe performance degradation under local optimization. To jointly achieve structured pruning and the memory efficiency of local pruning, we propose a divide-and-conquer strategy that decomposes the global pruning problem into coordinated subproblems across different modules, each of which fits within limited GPU memory. Building on this idea, we design \textbf{STRUPRUNE}, an ADMM-based framework that integrates structured sparsity into the pruning process, combining the memory efficiency of local pruning with the hardware compatibility of structured methods. We derive a closed-form analytical solution for structured pruning masks that provides an explicit rule for layer-wise sparsity allocation, and further develop an energy-based asymptotic framework yielding a softmax-form allocation scheme that simplifies optimization while adapting to heterogeneous layer importance. Experiments demonstrate that STRUPRUNE matches the perplexity of global structured pruning while reducing memory cost from $\mathcal{O}(N)$ to $\mathcal{O}(\sqrt{N})$, enabling practical deployment at the billion-parameter scale.

Foundation models (FMs) deployed in real-world tasks such as computer-use agents must integrate diverse modalities. How good are FMs at performing joint reasoning, simultaneously reasoning over multiple modalities, especially when the modalities interact and relate to each other to form cross-modal context? To better understand this problem, we study FMs on cross-modal conflicts: scenarios where conflicting evidence is presented across modalities. This allows us to examine whether FMs prioritize one modality over another or reason jointly to reconcile the conflict. Our experiments reveal that FMs can recognize conflicts in unimodal contexts, composed of a single modality, 90% of the time, but the ratio falls as low as 3% when evidence is split across modalities -- similar observations hold in cross-lingual contexts, composed of multiple languages. We trace this failure to cross-modal attention imbalance, showing that FMs exhibit extreme asymmetry in attention scores, disproportionately prioritizing certain modalities. We show that cross-modal attention imbalance does not go away by simply scaling up multimodal or multilingual datasets blindly, since they lack training examples that explicitly require cross-modal reasoning. We demonstrate that even a simple and scalable method of explicitly combining multiple modalities within each training instance significantly reduces attention imbalance. Reduced attention imbalance directly translates to improved downstream performance on several vision-language benchmarks. Our findings underscore the importance of systematically addressing cross-modal contexts to build reliable foundation models.

The impressive performance gains of modern language models currently rely on scaling parameters: larger models store more world knowledge and reason better. Yet compressing all world knowledge into parameters is unnecessary, as only a fraction is used per prompt, and impractical for edge devices with limited inference-time memory and compute. We address this shortcoming by a memory-augmented architecture and a pretraining strategy aligned with existing hardware paradigms. We introduce small language models that access large hierarchical parametric memory banks encoding world knowledge. During pretraining and inference, we fetch a small, context-dependent memory block and add it to the model. Our pretraining learns to store long-tail world knowledge in the memory parameters, while the small language model acts as an anchor capturing common knowledge and general reasoning abilities. Through trillion-token-scale experiments, we show significant gains: a 160M-parameters model augmented with an 18M-parameters memory fetched from a 4.6B memory bank obtains comparable performance to a regular model with more than 2x the parameters. Through extensive experiments, we study the optimal type and size of parametric memories in transformers, scaling them to over 21B parameters. We find that our proposed hierarchical feed-forward memories work robustly across transformer architectures, whether added during pretraining or post-hoc.