ReQuant: Fixed-Grid Discrete Refinement for Post-Training Quantization

Post-training quantization (PTQ) is widely used to reduce the memory and computational cost of large language models. Existing PTQ methods typically obtain an initial quantized model through heuristic rules or greedy optimization, and once quantization is completed the resulting integer assignments are usually treated as final. This observation motivates a complementary optimization stage within PTQ that keeps quantized weights improvable after an executable quantized model has been produced, while preserving the quantized format. We introduce ReQuant, a backpropagation-free fixed-grid refinement procedure for this stage. Agnostic to the PTQ initializer, ReQuant takes an existing quantized model as a feasible starting point and iteratively revisits its discrete weight assignments on the fixed quantization grid. Accepted updates strictly reduce the mean squared reconstruction error and remain on the original grid. In this way, ReQuant turns the initially fixed PTQ output into an iteratively optimizable discrete solution and serves as a plug-and-play post-processing stage for existing PTQ pipelines. Experiments across diverse model families, bit-widths, and downstream tasks show that ReQuant consistently improves quantized models from heterogeneous PTQ initializers, with especially large gains on simple initializers and lower bit-widths. Notably, ReQuant can refine a simple round-to-nearest initialization across multiple sweeps until it approaches or surpasses GPTAQ under the same quantization format. These results establish ReQuant as a practical complementary stage for further improving existing PTQ pipelines.

Beyond Rephrasing: Book-Level Organization Improves Synthetic Textbook Data for Mid-Training

Synthetic textbook data has improved language model pre-training, but prior work largely treats the benefit as a property of generated content or local rewriting style. We study a different factor: whether related content is organized into coherent book-level documents. We contribute both a scalable synthesis pipeline and controlled evidence that this organization matters. The pipeline retrieves source material from a pre-training corpus, clusters it into topical units, plans hierarchical tables of contents, and assembles source-grounded sections into complete books (our Full setting), yielding 686K textbooks (32B tokens) across 15,000+ disciplines. Replacing natural books in a mid-training mix with this corpus improves downstream performance by +1.09 on average. Controlled comparisons then disentangle the relevant design factors. A content-matched Split condition holds generated text and tokens fixed but treats each section as an independent document; Full's +1.02 mean gain isolates document packaging. A length-matched RandomConcat control that joins sections from different books remains below Full, ruling out document length alone. A retrieval-pool-matched Rephrase condition independently rewrites individual retrieved documents under the same audience-by-style scheme, without clustering, TOC planning, or book assembly; Full's +1.17 gain demonstrates the value of structured synthesis. On Llama3-8B, Full likewise outperforms both RandomConcat and Natural Books, supporting book-level organization as a useful axis for synthetic pre-training data design.

C^2GR: Coupled Comprehensive Generative Replay for a Continually Learnable Universal Segmentation Model

Universal segmentation models exhibit significant potential for diverse tasks involving different imaging modalities and segmentation objectives. Task-Incremental Learning provides a privacy-preserving approach to continually evolve a universal model on tasks from sequentially-arriving medical departments. However, training the model solely on the incoming task induces forgetting on past tasks, since consecutive tasks exhibit concurrent shifts in image appearance and segmentation objective. To address this problem, we propose a novel Coupled Comprehensive Generative Replay (C^2GR) framework that simultaneously synthesizes image-mask pairs of previous tasks to mitigate forgetting under concurrent appearance and objective shifts. This requires preserving image-mask correspondence for structure-realistic generation and bridging asynchronous optimization of the generator and segmentor for segmentation-oriented generation. Specifically, we propose a Bayesian Joint Diffusion (BJD) method that formulates the correspondence as conditional distributions optimized via conditional denoising. Furthermore, we develop a Relation-aware Unified Prompt Synchronization (RUPS) scheme to simultaneously modulate the generator and segmentor via a shared task-relation-aware prompt for synchronizing their optimization. Experiments on 20 tasks spanning diverse modalities and objectives demonstrate that C^2GR exhibits only a 2.44% drop in overall performance compared to joint training with all task data, effectively alleviating forgetting from the concurrent shifts. Our code will be made publicly available at https://github.com/mar-cry/C2GR.

MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection

Large language model agents increasingly rely on persistent memory to store past interactions, retrieve relevant demonstrations, and improve long-horizon task execution. However, this memory mechanism also creates a practical security vulnerability: an adversarial user may inject malicious records into the agent's memory through ordinary interaction, and these records can later be retrieved to steer the agent's reasoning and actions. Existing defenses primarily focus on online intervention, such as prompt filtering or output blocking, but they do not address the post-hoc question of which stored memories are responsible after harmful behavior has already been observed. We propose \textbf{MemAudit}, a post-hoc causal memory auditing framework for memory-augmented LLM agents. The framework combines two complementary signals: (1) a counterfactual memory influence score that measures each memory's causal contribution to harmful outputs, and (2) a memory consistency graph that identifies structurally anomalous memories within the broader memory store. We evaluate MemAudit against MINJA, a query-only memory injection attack in which malicious records are generated and stored through normal agent interactions rather than direct memory-bank modification. Across both QA and reasoning-agent settings, MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios. The results show that QA attack success is reduced from $70\%$ to $0\%$, while RAP attack success drops from $83.3\%$ to $0\%$.

NeuroWeaver: An Autonomous Evolutionary Agent for Exploring the Programmatic Space of EEG Analysis Pipelines

Although foundation models have demonstrated remarkable success in general domains, the application of these models to electroencephalography (EEG) analysis is constrained by substantial data requirements and high parameterization. These factors incur prohibitive computational costs, thereby impeding deployment in resource-constrained clinical environments. Conversely, general-purpose automated machine learning frameworks are often ill-suited for this domain, as exploration within an unbounded programmatic space fails to incorporate essential neurophysiological priors and frequently yields solutions that lack scientific plausibility. To address these limitations, we propose NeuroWeaver, a unified autonomous evolutionary agent designed to generalize across diverse EEG datasets and tasks by reformulating pipeline engineering as a discrete constrained optimization problem. Specifically, we employ a Domain-Informed Subspace Initialization to confine the search to neuroscientifically plausible manifolds, coupled with a Multi-Objective Evolutionary Optimization that dynamically balances performance, novelty, and efficiency via self-reflective refinement. Empirical evaluations across five heterogeneous benchmarks demonstrate that NeuroWeaver synthesizes lightweight solutions that consistently outperform state-of-the-art task-specific methods and achieve performance comparable to large-scale foundation models, despite utilizing significantly fewer parameters.

HESTIA: A Hessian-Guided Differentiable Quantization-Aware Training Framework for Extremely Low-Bit LLMs

As large language models (LLMs) continue to scale, deployment is increasingly bottlenecked by the memory wall, motivating a shift toward extremely low-bit quantization. However, most quantization-aware training (QAT) methods apply hard rounding and the straight-through estimator (STE) from the beginning of the training, which prematurely discretizes the optimization landscape and induces persistent gradient mismatch between latent weights and quantized weights, hindering effective optimization of quantized models. To address this, we propose Hestia, a Hessian-guided differentiable QAT framework for extremely low-bit LLMs, which replaces the rigid step function with a temperature-controlled softmax relaxation to maintain gradient flow early in training while progressively hardening quantization. Furthermore, Hestia leverages a tensor-wise Hessian trace metric as a lightweight curvature signal to drive fine-grained temperature annealing, enabling sensitivity-aware discretization across the model. Evaluations on Llama-3.2 show that Hestia consistently outperforms existing ternary QAT baselines, yielding average zero-shot improvements of 5.39% and 4.34% for the 1B and 3B models. These results indicate that Hessian-guided relaxation effectively recovers representational capacity, establishing a more robust training path for 1.58-bit LLMs. The code is available at https://github.com/hestia2026/Hestia.

MiMo-V2-Flash Technical Report

We present MiMo-V2-Flash, a Mixture-of-Experts (MoE) model with 309B total parameters and 15B active parameters, designed for fast, strong reasoning and agentic capabilities. MiMo-V2-Flash adopts a hybrid attention architecture that interleaves Sliding Window Attention (SWA) with global attention, with a 128-token sliding window under a 5:1 hybrid ratio. The model is pre-trained on 27 trillion tokens with Multi-Token Prediction (MTP), employing a native 32k context length and subsequently extended to 256k. To efficiently scale post-training compute, MiMo-V2-Flash introduces a novel Multi-Teacher On-Policy Distillation (MOPD) paradigm. In this framework, domain-specialized teachers (e.g., trained via large-scale reinforcement learning) provide dense and token-level reward, enabling the student model to perfectly master teacher expertise. MiMo-V2-Flash rivals top-tier open-weight models such as DeepSeek-V3.2 and Kimi-K2, despite using only 1/2 and 1/3 of their total parameters, respectively. During inference, by repurposing MTP as a draft model for speculative decoding, MiMo-V2-Flash achieves up to 3.6 acceptance length and 2.6x decoding speedup with three MTP layers. We open-source both the model weights and the three-layer MTP weights to foster open research and community collaboration.

MiMo-Audio: Audio Language Models are Few-Shot Learners

Existing audio language models typically rely on task-specific fine-tuning to accomplish particular audio tasks. In contrast, humans are able to generalize to new audio tasks with only a few examples or simple instructions. GPT-3 has shown that scaling next-token prediction pretraining enables strong generalization capabilities in text, and we believe this paradigm is equally applicable to the audio domain. By scaling MiMo-Audio's pretraining data to over one hundred million of hours, we observe the emergence of few-shot learning capabilities across a diverse set of audio tasks. We develop a systematic evaluation of these capabilities and find that MiMo-Audio-7B-Base achieves SOTA performance on both speech intelligence and audio understanding benchmarks among open-source models. Beyond standard metrics, MiMo-Audio-7B-Base generalizes to tasks absent from its training data, such as voice conversion, style transfer, and speech editing. MiMo-Audio-7B-Base also demonstrates powerful speech continuation capabilities, capable of generating highly realistic talk shows, recitations, livestreaming and debates. At the post-training stage, we curate a diverse instruction-tuning corpus and introduce thinking mechanisms into both audio understanding and generation. MiMo-Audio-7B-Instruct achieves open-source SOTA on audio understanding benchmarks (MMSU, MMAU, MMAR, MMAU-Pro), spoken dialogue benchmarks (Big Bench Audio, MultiChallenge Audio) and instruct-TTS evaluations, approaching or surpassing closed-source models. Model checkpoints and full evaluation suite are available at https://github.com/XiaomiMiMo/MiMo-Audio.

Fairy$\pm i$: the First 2-bit Complex LLM with All Parameters in $\{\pm1, \pm i\}$

Quantization-Aware Training (QAT) integrates quantization into the training loop, enabling LLMs to learn robust low-bit representations, and is widely recognized as one of the most promising research directions. All current QAT research focuses on minimizing quantization error on full-precision models, where the full-precision accuracy acts as an upper bound (accuracy ceiling). No existing method has even attempted to surpass this ceiling. To break this ceiling, we propose a new paradigm: raising the ceiling (full-precision model), and then still quantizing it efficiently into 2 bits. We propose Fairy$\pm i$, the first 2-bit quantization framework for complex-valued LLMs. Specifically, our method leverages the representational advantages of the complex domain to boost full-precision accuracy. We map weights to the fourth roots of unity $\{\pm1, \pm i\}$, forming a perfectly symmetric and information-theoretically optimal 2-bit representation. Importantly, each quantized weight has either a zero real or imaginary part, enabling multiplication-free inference using only additions and element swaps. Experimental results show that Fairy$\pm i$ outperforms the ceiling of existing 2-bit quantization approaches in terms of both PPL and downstream tasks, while maintaining strict storage and compute efficiency. This work opens a new direction for building highly accurate and practical LLMs under extremely low-bit constraints.

F^2TTA: Free-Form Test-Time Adaptation on Cross-Domain Medical Image Classification via Image-Level Disentangled Prompt Tuning

Test-Time Adaptation (TTA) has emerged as a promising solution for adapting a source model to unseen medical sites using unlabeled test data, due to the high cost of data annotation. Existing TTA methods consider scenarios where data from one or multiple domains arrives in complete domain units. However, in clinical practice, data usually arrives in domain fragments of arbitrary lengths and in random arrival orders, due to resource constraints and patient variability. This paper investigates a practical Free-Form Test-Time Adaptation (F$^{2}$TTA) task, where a source model is adapted to such free-form domain fragments, with shifts occurring between fragments unpredictably. In this setting, these shifts could distort the adaptation process. To address this problem, we propose a novel Image-level Disentangled Prompt Tuning (I-DiPT) framework. I-DiPT employs an image-invariant prompt to explore domain-invariant representations for mitigating the unpredictable shifts, and an image-specific prompt to adapt the source model to each test image from the incoming fragments. The prompts may suffer from insufficient knowledge representation since only one image is available for training. To overcome this limitation, we first introduce Uncertainty-oriented Masking (UoM), which encourages the prompts to extract sufficient information from the incoming image via masked consistency learning driven by the uncertainty of the source model representations. Then, we further propose a Parallel Graph Distillation (PGD) method that reuses knowledge from historical image-specific and image-invariant prompts through parallel graph networks. Experiments on breast cancer and glaucoma classification demonstrate the superiority of our method over existing TTA approaches in F$^{2}$TTA. Code is available at https://github.com/mar-cry/F2TTA.