A safety direction estimated in a source LLM is transported to a target generator through lightweight alignment on benign data alone, matching native safety performance without any target-side unsafe data.
hub
Unimax: Fairer and more effective language sampling for large-scale multilingual pretraining
18 Pith papers cite this work. Polarity classification is still indexing.
18
Pith papers citing it
abstract
Pretrained multilingual large language models have typically used heuristic temperature-based sampling to balance between different languages. However previous work has not systematically evaluated the efficacy of different pretraining language distributions across model scales. In this paper, we propose a new sampling method, UniMax, that delivers more uniform coverage of head languages while mitigating overfitting on tail languages by explicitly capping the number of repeats over each language's corpus. We perform an extensive series of ablations testing a range of sampling strategies on a suite of multilingual benchmarks, while varying model scale. We find that UniMax outperforms standard temperature-based sampling, and the benefits persist as scale increases. As part of our contribution, we release: (i) an improved and refreshed mC4 multilingual corpus consisting of 29 trillion characters across 107 languages, and (ii) a suite of pretrained umT5 model checkpoints trained with UniMax sampling.
hub tools
citation-role summary
background 3
citation-polarity summary
roles
background 3polarities
background 3representative citing papers
DanceCrafter generates high-fidelity, text-controlled dance sequences using a new Choreographic Syntax framework and a large fine-grained motion dataset.
A video foundation model trained on human demonstrations generates zero-shot plans that convert to executable robot actions on novel scenes and tasks.
ASTRA is a plug-and-play training-free method for precise multi-subject video editing that uses prompt-guided multimodal alignment and prior-based mask retargeting to avoid attention dilution and boundary issues.
Fine-tuning a video generator on a new 17K reasoning-video dataset improves Chain-of-Frame reasoning, and adding learnable visual/textual reasoning tokens yields further gains on external benchmarks.
A training-free 3D scene-adaptive human animation framework that controls human motion and camera trajectories via ground-adaptive retargeting and visibility-masked point-cloud fusion in a diffusion backbone.
ActWorld extends navigation-centric world models to support mid-rollout object interactions via chunk-autoregressive generation, action-aware memory routing, and a persistent memory bank, backed by a 100K annotated interaction dataset.
Modality Forcing lets a single DiT produce image and depth outputs in any order after training on sparse real-world depth, with larger image-pretrained models yielding better depth accuracy and a 57% AbsRel reduction versus prior joint generative baselines.
Transfer-aware data allocation derived from observed power-law scaling laws for asymmetric knowledge transfer in 3D medical imaging outperforms standard proportional sampling by up to 58% and generalizes to new budgets.
Uni-ViGU unifies video generation and understanding by extending a diffusion video generator with unified continuous-discrete flow matching, modality-driven MoE layers, and bidirectional training stages that repurpose generative knowledge for discriminative tasks.
Rolling Sink is a training-free cache adjustment technique that maintains visual consistency in autoregressive video diffusion models for ultra-long open-ended generation beyond training horizons.
AstraNav-World unifies diffusion video generation and vision-language action planning in a single bidirectional model that improves trajectory accuracy, success rates, and zero-shot real-world adaptation in embodied navigation.
KeyTailor improves video virtual try-on realism by using instruction-guided keyframes to enhance garment details and background integrity in DiT models without major architectural changes.
SkyReels-V2 produces infinite-length film videos via MLLM-based captioning, progressive pretraining, motion RL, and diffusion forcing with non-decreasing noise schedules.
DomainShuttle introduces domain-aware modeling and token separation techniques to achieve high subject fidelity with generative flexibility in open-domain subject-driven text-to-video tasks.
Omni-Customizer proposes an end-to-end framework using Omni-Context Fusion, Masked TTS Cross-Attention, Semantic-Anchored Multimodal RoPE, and specialized training curricula to achieve precise multimodal identity binding in joint audio-video generation.
Wan releases open 1.3B and 14B video diffusion models claiming superior performance over open-source and commercial baselines across multiple tasks with consumer-grade efficiency.
A survey that taxonomizes data mixing strategies for LLM pretraining into static rule-based, learning-based, and dynamic adaptive families while highlighting transferability challenges and evaluation gaps.
citing papers explorer
-
Do Models Share Safety Representations? Cross-Model Steering for Safe Visual Generation
A safety direction estimated in a source LLM is transported to a target generator through lightweight alignment on benign data alone, matching native safety performance without any target-side unsafe data.
-
DanceCrafter: Fine-Grained Text-Driven Controllable Dance Generation via Choreographic Syntax
DanceCrafter generates high-fidelity, text-controlled dance sequences using a new Choreographic Syntax framework and a large fine-grained motion dataset.
-
Large Video Planner Enables Generalizable Robot Control
A video foundation model trained on human demonstrations generates zero-shot plans that convert to executable robot actions on novel scenes and tasks.
-
ASTRA: Let Arbitrary Subjects Transform in Video Editing
ASTRA is a plug-and-play training-free method for precise multi-subject video editing that uses prompt-guided multimodal alignment and prior-based mask retargeting to avoid attention dilution and boundary issues.
-
OpenCoF: Learning to Reason Through Video Generation
Fine-tuning a video generator on a new 17K reasoning-video dataset improves Chain-of-Frame reasoning, and adding learnable visual/textual reasoning tokens yields further gains on external benchmarks.
-
3D Scene-Adaptive Trajectory-Controllable Human Image Animation with Camera Movement
A training-free 3D scene-adaptive human animation framework that controls human motion and camera trajectories via ground-adaptive retargeting and visibility-masked point-cloud fusion in a diffusion backbone.
-
ActWorld: From Explorable to Interactive World Model via Action-Aware Memory
ActWorld extends navigation-centric world models to support mid-rollout object interactions via chunk-autoregressive generation, action-aware memory routing, and a persistent memory bank, backed by a 100K annotated interaction dataset.
-
Modality Forcing for Scalable Spatial Generation
Modality Forcing lets a single DiT produce image and depth outputs in any order after training on sparse real-world depth, with larger image-pretrained models yielding better depth accuracy and a 57% AbsRel reduction versus prior joint generative baselines.
-
Knowledge Transfer Scaling Laws for 3D Medical Imaging
Transfer-aware data allocation derived from observed power-law scaling laws for asymmetric knowledge transfer in 3D medical imaging outperforms standard proportional sampling by up to 58% and generalizes to new budgets.
-
Uni-ViGU: Towards Unified Video Generation and Understanding via A Diffusion-Based Video Generator
Uni-ViGU unifies video generation and understanding by extending a diffusion video generator with unified continuous-discrete flow matching, modality-driven MoE layers, and bidirectional training stages that repurpose generative knowledge for discriminative tasks.
-
Rolling Sink: Bridging Limited-Horizon Training and Open-Ended Testing in Autoregressive Video Diffusion
Rolling Sink is a training-free cache adjustment technique that maintains visual consistency in autoregressive video diffusion models for ultra-long open-ended generation beyond training horizons.
-
AstraNav-World: World Model for Foresight Control and Consistency
AstraNav-World unifies diffusion video generation and vision-language action planning in a single bidirectional model that improves trajectory accuracy, success rates, and zero-shot real-world adaptation in embodied navigation.
-
The devil is in the details: Enhancing Video Virtual Try-On via Keyframe-Driven Details Injection
KeyTailor improves video virtual try-on realism by using instruction-guided keyframes to enhance garment details and background integrity in DiT models without major architectural changes.
-
SkyReels-V2: Infinite-length Film Generative Model
SkyReels-V2 produces infinite-length film videos via MLLM-based captioning, progressive pretraining, motion RL, and diffusion forcing with non-decreasing noise schedules.
-
DomainShuttle: Freeform Open Domain Subject-driven Text-to-video Generation
DomainShuttle introduces domain-aware modeling and token separation techniques to achieve high subject fidelity with generative flexibility in open-domain subject-driven text-to-video tasks.
-
Omni-Customizer: End-to-End MultiModal Customization for Joint Audio-Video Generation
Omni-Customizer proposes an end-to-end framework using Omni-Context Fusion, Masked TTS Cross-Attention, Semantic-Anchored Multimodal RoPE, and specialized training curricula to achieve precise multimodal identity binding in joint audio-video generation.
-
Wan: Open and Advanced Large-Scale Video Generative Models
Wan releases open 1.3B and 14B video diffusion models claiming superior performance over open-source and commercial baselines across multiple tasks with consumer-grade efficiency.
-
Data Mixing for Large Language Models Pretraining: A Survey and Outlook
A survey that taxonomizes data mixing strategies for LLM pretraining into static rule-based, learning-based, and dynamic adaptive families while highlighting transferability challenges and evaluation gaps.