Provides a domain-agnostic world-modeling framework that factorizes latent targets into orthogonal predictive components, with dedicated prediction branches and synthesis for multi-domain forecasting and intervention. Demonstrates improved dynamics and long-horizon rollouts across seven domains and includes experimental biological validation.
Turns implemented functionality in open-source codebases into executable RL training environments by automatically generating behavioral specifications, execution-grounded tests, and post-rollout filtering; yields 5,545 verifiable coding tasks across 23 languages to train coding agents.
Provides mixed-domain, verifiable RL training environments for LLM agents (code, cyber, knowledge work, web dev, music) as Parquet datasets, with domain-specific verifiers, Docker artifacts and links to training code for reproducible agentic RL experiments.
Open-weight 309B Mixture-of-Experts causal LLM with 15.5B active parameters and a native 1M-token context for coding and AI R&D. Combines Sliding-Window Attention and DeepSeek Sparse Attention (no full-attention layers), supports FP8 inference; weights under MIT license.
Learns dynamics-aligned latent representations for parametric PDE forecasting using masked-latent predictive pretraining, a geometry projector that aligns latent trajectory geometry, and a physics-structured latent predictor that separates shared evolution from parameter-dependent responses.
Improves test-time scaling of looped transformers by adaptively assigning extra recurrent iterations to tokens that benefit most. TaH2 is a post-training method that jointly trains an iteration decider with the backbone using lookahead depth supervision, boosting the accuracy–compute slope and peak accuracy on challenging benchmarks.
Compresses recurrent states in linear-attention LLMs via spatial–temporal post-training quantization, allocating bits by error lifetime and per-row impact to preserve accuracy (6-bit ≈ FP32) while cutting serving memory up to 68.7%.
Demonstrates that additive weighted aggregation in rubric-based reinforcement learning enables reward hacking—models improve rubric coverage but produce worse answers—and proposes ProRubric, an offline regrouping that aggregates checklist criteria into conjunctive protocol-level dimensions to raise appropriateness without losing coverage.
Analyzes why self-evolving reasoning models collapse under repeated self-training and proposes R-Quest: a feedback-driven pipeline that trains solvers to reject invalid questions and uses a frozen base model to detect task-level repetition, filtering training data to sustain multi-round gains.
Generates 5-second text- or image-conditioned videos with synchronized 44 kHz audio (including lip-sync) and built-in super-resolution to 1920×1080; available in Lite (3B) and Pro (29B) variants with code and checkpoints released under an MIT license.
Analyzes cross-tokenizer on-policy distillation (OPD) and shows that focusing supervision on strictly aligned token positions—or a student-selected top-16 subset of the shared vocabulary—retains most distillation gains, while adding span-level MSE supervision reduces downstream accuracy.
Text-only sparse Mixture-of-Experts language model for long-context reasoning, coding, and agentic tool use; 501B total / ~23B active parameters, 1M-token context, and BF16/FP8/NVFP4 checkpoints released under Apache 2.0.