Enfold: Folding World Model Imagination into Predictive Representations for Ultra-Efficient Embodied Control

World generative models are typically used through what they produce: a rendered future, a video-conditioned action, or latent context computed by a costly generative branch. We argue that their more reusable asset is the computation that constructs a future. As a generator transforms a corrupted future into a coherent trajectory, its intermediate states organize appearance, spatial layout, and interaction across levels of abstraction. Can this future-generative computation be internalized in a representation inferred from the present alone? We present Enfold, which transfers this computation into a representation predicted from the current visual context and language instruction. During training, multi-level states exposed as the generator processes the observed future supervise a current-only encoder. The learned representation is fed back to condition future generation and is read by task heads without allowing task gradients to reshape the encoder. At deployment, action prediction no longer executes the generator. Across LIBERO, RoboTwin2.0, and real-robot tasks, Enfold supports strong control while reducing action latency by $3.7\times$ relative to Fast--WAM, Enfold-Flash reaches $10.1\times$. Representation analyses show that it suppresses nuisance variation and preferentially captures changes that emerge over longer horizons. When the current scene is altered by human intervention, both the generated continuation and the executed actions adapt, which is inconsistent with fixed trajectory replay. These results recast a world generator as a source of predictive control representations: its future need not be materialized at every step if its internal structure can be enfolded into the present.

Enfold: Folding World-Generator Computation into Predictive Representations for Efficient Embodied Control

World generative models are typically used through what they produce: a rendered future, a video-conditioned action, or latent context computed by a costly generative branch. We argue that their more reusable asset is the computation that constructs a future. As a generator transforms a corrupted future into a coherent trajectory, its intermediate states organize appearance, spatial layout, and interaction across levels of abstraction. Can this future-generative computation be internalized in a representation inferred from the present alone? We present Enfold, which transfers this computation into a representation predicted from the current visual context and language instruction. During training, multi-level states exposed as the generator processes the observed future supervise a current-only encoder. The learned representation is fed back to condition future generation and is read by task heads without allowing task gradients to reshape the encoder. At deployment, action prediction no longer executes the generator. Across LIBERO, RoboTwin2.0, and real-robot tasks, Enfold supports strong control while reducing action latency by $3.7\times$ relative to Fast--WAM, Enfold-Flash reaches $10.1\times$. Representation analyses show that it suppresses nuisance variation and preferentially captures changes that emerge over longer horizons. When the current scene is altered by human intervention, both the generated continuation and the executed actions adapt, which is inconsistent with fixed trajectory replay. These results recast a world generator as a source of predictive control representations: its future need not be materialized at every step if its internal structure can be enfolded into the present.

Joint Text-Audio Alignment for EEG-to-Text Decoding in Chinese Speech Production and Perception

Decoding speech information directly from scalp electroencephalography (EEG) into text provides a potential non-invasive neural communication pathway for individuals with severe speech and motor impairments. Compared with invasive approaches such as electrocorticography, EEG is safer and more widely deployable, yet substantially more challenging to decode.This challenge is exacerbated for Chinese sentence decoding, which must handle a high-dimensional output space with thousands of characters, severe inter-subject variability, and low signal-to-noise ratios for text alignment.Existing methods commit to a single supervisory axis---either text semantics or audio acoustic features---yet neither can simultaneously satisfy the demands of sentence-level discriminability and fine-grained temporal resolution required for large-vocabulary Chinese decoding. We introduce EEGAlign, a novel parameter-efficient framework that jointly aligns EEG with two axes---text alignment with BGE-M3 text embeddings and audio alignment with wav2vec~2.0 speech features via contrastive learning followed by CTC character-sequence decoding. On ChineseEEG-2 data, EEGAlign yields state-of-the-art closed-set sentence classification performance, reaching up to 82.37% Top-1 accuracy on Reading Aloud EEG and 41.43% on Passive Listening EEG out of 101 candidates. Ablation studies show that the two alignment axes are highly complementary: combining them yields consistently better performance than either alone. To the best of our knowledge, this is the first study on decoding large-vocabulary Chinese sentences from non-invasive EEG during overt speech production, and achieving strong classification performance with relatively large closed-set candidate-sentence setting.

CLAP: Closed-Loop Training, Evaluation, and Release Control for Domain Agent Post-training

Domain agents often face noisy business data, uncertain post-training gains, offline/application mismatch, and adapter-release risk. This paper presents CLAP (Closed-Loop Agent Post-training), a closed-loop method that converts business data into structured SFT samples, decision-preference samples, holdout sets, risk diagnostics, and release-gate records. CLAP combines data validation, target/evidence normalization, reward/KL diagnosis, offline gates, and application-chain replay to decide whether an adapter is suitable for the target application chain. On five anonymized manufacturing-scenario batches, QLoRA-style LoRA-SFT yields modest average gains: overall score increases by 0.0098, pass rate by 0.0240, and evidence accuracy by 0.0280, while hallucination and wrong facts decrease. Yet only 3 of 5 batches improve, some batches regress, and GRPO exposes high KL risks. Application-chain replay further shows that RAG is necessary for factual extraction; under the same 3B backbone and 100 replay cases, an application-RAG-oriented LoRA-SFT adapter improves value, core fields, and answer-evidence doc/page matching over base+RAG, but increases latency. These results support managing domain-agent post-training through an integrated data-training-evaluation-release loop rather than relying on training completion or a single offline score.

CoMo3R-SLAM: Collaborative Monocular Dense SLAM with Learned 3D Reconstruction Priors for Outdoor Multi-Agent Systems

Collaborative dense SLAM is essential for multi-robot teams to achieve scalable and consistent 3D perception across large-scale outdoor environments. Existing systems typically depend on depth sensors, incurring significant payload, power, and calibration costs. Monocular RGB cameras are a lightweight alternative, but collaborative monocular dense SLAM remains difficult due to scale ambiguity, unreliable inter-agent data association, especially in outdoor scenes where low overlap and repetitive structures make traditional feature matching unreliable, motivating robust geometric information. We propose CoMo3R-SLAM, the first collaborative monocular dense RGB SLAM system that leverages robust learned feed-forward 3D reconstruction priors for outdoor multi-agent mapping. Each agent runs a prior-guided front-end for real-time tracking and local dense fusion, while a coordinator performs dense pointmap matching for cross-agent verification, closed-form Sim(3) gauge synchronization, and GPU-accelerated global bundle adjustment with segment-level depth optimization. Requiring neither depth sensors nor parametric intrinsics, our system produces robust cross-agent constraints and globally consistent metric maps from monocular RGB alone. On Tanks and Temples and Waymo sequences, CoMo3R-SLAM achieves the best ATE on three of four Tanks and Temples scenes and competitive Waymo accuracy, matching or exceeding state-of-the-art RGB-D methods while running online at 8 FPS.

StepAudio 2.5 Technical Report

Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.

HFP-SAM: Hierarchical Frequency Prompted SAM for Efficient Marine Animal Segmentation

Marine Animal Segmentation (MAS) aims at identifying and segmenting marine animals from complex marine environments. Most of previous deep learning-based MAS methods struggle with the long-distance modeling issue. Recently, Segment Anything Model (SAM) has gained popularity in general image segmentation. However, it lacks of perceiving fine-grained details and frequency information. To this end, we propose a novel learning framework, named Hierarchical Frequency Prompted SAM (HFP-SAM) for high-performance MAS. First, we design a Frequency Guided Adapter (FGA) to efficiently inject marine scene information into the frozen SAM backbone through frequency domain prior masks. Additionally, we introduce a Frequency-aware Point Selection (FPS) to generate highlighted regions through frequency analysis. These regions are combined with the coarse predictions of SAM to generate point prompts and integrate into SAM's decoder for fine predictions. Finally, to obtain comprehensive segmentation masks, we introduce a Full-View Mamba (FVM) to efficiently extract spatial and channel contextual information with linear computational complexity. Extensive experiments on four public datasets demonstrate the superior performance of our approach. The source code is publicly available at https://github.com/Drchip61/TIP-HFP-SAM.

DMESR: Dual-view MLLM-based Enhancing Framework for Multimodal Sequential Recommendation

Sequential Recommender Systems (SRS) aim to predict users' next interaction based on their historical behaviors, while still facing the challenge of data sparsity. With the rapid advancement of Multimodal Large Language Models (MLLMs), leveraging their multimodal understanding capabilities to enrich item semantic representation has emerged as an effective enhancement strategy for SRS. However, existing MLLM-enhanced recommendation methods still suffer from two key limitations. First, they struggle to effectively align multimodal representations, leading to suboptimal utilization of semantic information across modalities. Second, they often overly rely on MLLM-generated content while overlooking the fine-grained semantic cues contained in the original textual data of items. To address these issues, we propose a Dual-view MLLM-based Enhancing framework for multimodal Sequential Recommendation (DMESR). For the misalignment issue, we employ a contrastive learning mechanism to align the cross-modal semantic representations generated by MLLMs. For the loss of fine-grained semantics, we introduce a cross-attention fusion module that integrates the coarse-grained semantic knowledge obtained from MLLMs with the fine-grained original textual semantics. Finally, these two fused representations can be seamlessly integrated into the downstream sequential recommendation models. Extensive experiments conducted on three real-world datasets and three popular sequential recommendation architectures demonstrate the superior effectiveness and generalizability of our proposed approach.

Interactive Spatial-Frequency Fusion Mamba for Multi-Modal Image Fusion

Multi-Modal Image Fusion (MMIF) aims to combine images from different modalities to produce fused images, retaining texture details and preserving significant information. Recently, some MMIF methods incorporate frequency domain information to enhance spatial features. However, these methods typically rely on simple serial or parallel spatial-frequency fusion without interaction. In this paper, we propose a novel Interactive Spatial-Frequency Fusion Mamba (ISFM) framework for MMIF. Specifically, we begin with a Modality-Specific Extractor (MSE) to extract features from different modalities. It models long-range dependencies across the image with linear computational complexity. To effectively leverage frequency information, we then propose a Multi-scale Frequency Fusion (MFF). It adaptively integrates low-frequency and high-frequency components across multiple scales, enabling robust representations of frequency features. More importantly, we further propose an Interactive Spatial-Frequency Fusion (ISF). It incorporates frequency features to guide spatial features across modalities, enhancing complementary representations. Extensive experiments are conducted on six MMIF datasets. The experimental results demonstrate that our ISFM can achieve better performances than other state-of-the-art methods. The source code is available at https://github.com/Namn23/ISFM.

Training LLMs with Fault Tolerant HSDP on 100,000 GPUs

Large-scale training systems typically use synchronous training, requiring all GPUs to be healthy simultaneously. In our experience training on O(100K) GPUs, synchronous training results in a low efficiency due to frequent failures and long recovery time. To address this problem, we propose a novel training paradigm, Fault Tolerant Hybrid-Shared Data Parallelism (FT-HSDP). FT-HSDP uses data parallel replicas as units of fault tolerance. When failures occur, only a single data-parallel replica containing the failed GPU or server is taken offline and restarted, while the other replicas continue training. To realize this idea at scale, FT-HSDP incorporates several techniques: 1) We introduce a Fault Tolerant All Reduce (FTAR) protocol for gradient exchange across data parallel replicas. FTAR relies on the CPU to drive the complex control logic for tasks like adding or removing participants dynamically, and relies on GPU to perform data transfer for best performance. 2) We introduce a non-blocking catch-up protocol, allowing a recovering replica to join training with minimal stall. Compared with fully synchronous training at O(100K) GPUs, FT-HSDP can reduce the stall time due to failure recovery from 10 minutes to 3 minutes, increasing effective training time from 44\% to 80\%. We further demonstrate that FT-HSDP's asynchronous recovery does not bring any meaning degradation to the accuracy of the result model.