探索全球前沿学术脉络

Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning

We introduce Nemotron 3 Ultra, a 550 billion total and 55 billion active parameter Mixture-of-Experts Hybrid Mamba-Attention language model. We pre-trained Nemotron 3 Ultra on 20 trillion text tokens, then extended the context length to 1M tokens, and post-trained using Supervised Fine Tuning (SFT), Reinforcement Learning (RL), and Multi-teacher On-Policy Distillation (MOPD). Nemotron 3 Ultra is our most capable model yet, employing multiple key technologies - LatentMoE, Multi Token Prediction (MTP), NVFP4 pre-training, multi-environment RLVR, MOPD, and reasoning budget control. Nemotron 3 Ultra achieves up to ~6x higher inference throughput as compared to state-of-the-art publicly available LLMs while attaining on-par accuracy. The state-of-the-art accuracy, high inference throughput, and 1M token context length make Nemotron 3 Ultra ideal for long-running autonomous agentic tasks. We open-source the base, post-trained, and quantized checkpoints, along with the training data and recipe on HuggingFace.

Light Interaction: Training-Free Inference Acceleration for Interactive Video World Models

arXiv:2605.31158v3 Announce Type: replace-cross Abstract: Interactive video world models generate video chunk by chunk in response to user-controlled camera movements, enabling applications such as real-time game simulation, virtual scene navigation, and embodied AI training. However, scaling to long interactive trajectories is prohibitively expensive due to growing context memory, quadratic attention complexity, and repeated denoising steps. We present Light Interaction, a training-free inference acceleration framework for interactive video world models. Our key insight is that interaction naturally enables trajectory-dependent adaptive computation: retrieved spatial memory can be discarded during novel exploration, temporal context can be adjusted according to local latent dynamics, and early-step model outputs can be reused when the camera revisits familiar regions. Based on this insight, Light Interaction combines adaptive context management, denoising cache acceleration, and hardware-software co-designed 3D block sparse attention with fused Triton kernels. Evaluated on HY-WorldPlay and Matrix-Game-3.0, Light Interaction achieves up to 2.59x speedup without model retraining while maintaining competitive visual quality.

MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling

We present MaxProof, a population-level test-time scaling framework for competition-level mathematical proof in the MiniMax-M3 series. M3 first trains three proof-oriented capabilities – proof generation, proof verification, and critique-conditioned proof repair – using a defense-in-depth generative verifier engineered for low false-positive rate. These capabilities are merged into a single released M3 model. At test time, MaxProof treats the model as a generator, verifier, refiner, and ranker, searches over a population of candidate proofs, and returns one final proof through tournament selection. With MaxProof test-time scaling, the M3 model reaches 35/42 on IMO 2025 and 36/42 on USAMO 2026, exceeding the human gold-medal threshold on both.

Mitigating scalability challenges in LUT-based neural networks via pruning optimisations

arXiv:2407.02362v3 Announce Type: replace-cross Abstract: Modern deep neural networks heavily rely on a large number of multiply-accumulate operations, which constitute the predominant computational cost. To address this, Look-Up Table (LUT)-based matrix multiplications have emerged as a promising alternative for reducing the computational cost and time of the multiply-accumulate operations in a neural network. However, the LUT-based neural network still faces the scalability challenge due to the inherent limitations of LUT-based matrix multiplication. To mitigate these scalability limitations, this paper proposes a scalable and energy-efficient LUT-based approximate matrix multiplication unit (LUT-MU) constituting the basic component of the neural networks by integrating a pruning strategy on the MADDNESS algorithm, a LUT-based matrix multiplication methodology. With increasing problem size and precision demands in matrix multiplication, our proposed LUT-MU architecture effectively constrains resource expansion. The case study shows that deploying our LUT-MU in neural network architectures, including fully connected layers (MNIST) and ResNets (CIFAR-10, ImageNet)-on XCZU7EV and XCZU19EG FPGAs, produces up to $1.6 \times$ throughput improvement and $4.2 \times$ energy efficiency gains over mainstream CUDA-based network implementations, and $1.8\times$ energy efficiency compared to leading quantised neural network implementations, with moderate impact on accuracy. Compared to original MADDNESS-based neural networks, our LUT-MU shows $1.3$ to $2.6\times$ resource savings based on various resolution configuration settings of MADDNESS.