Gated DeltaNet-2: Decoupling Erase and Write Gates in Linear Attention
Gated DeltaNet-2 is a new linear attention architecture from NVIDIA Labs that separates the erase and write operations in the delta-rule update into independent channel-wise gates, generalizing both Gated DeltaNet and Kimi Delta Attention (KDA). The model introduces a chunkwise WY algorithm with channel-wise decay and a gate-aware backward pass for efficient parallel training. At 1.3B parameters trained on 100B FineWeb-Edu tokens, it outperforms Mamba-2, Gated DeltaNet, KDA, and Mamba-3 variants on language modeling, commonsense reasoning, and long-context RULER needle-in-a-haystack retrieval benchmarks. Code is publicly released via NVlabs on GitHub.
Related guides (3)
Related events (8)
Comparative study finds xLSTM outperforms Mamba-2 and Gated DeltaNet on complex sequence tasks
A new arXiv paper compares three subquadratic sequence modeling architectures — xLSTM, Mamba-2, and Gated DeltaNet — across code model pre-training, LLM distillation, and time-series foundation model pre-training. xLSTM consistently delivers the strongest performance, which the authors attribute to more flexible and stable memory correction via its gating scheme. The paper provides a unified formulation and analysis of state tracking and memory dynamics across the three architectures, with corroborating results on synthetic length-generalization tasks.
AGDO: Attention-guided denoising and optimization framework improves diffusion language model reasoning
Researchers propose AGDO, a framework that replaces random masking in diffusion large language models (dLLMs) with attention-guided denoising order and token weighting during fine-tuning and reinforcement learning. The work is motivated by an empirical finding that tokens with stronger attention to unmasked context are more stable and critical for reasoning. Experiments on math and coding benchmarks show AGDO outperforms existing post-training methods for dLLMs, advancing the case for attention-aware training in parallel-decoding language models.
Test-Time Training End-to-End (TTT-E2E) Retrains Model Weights to Handle Long Inputs
Researchers from Astera Institute, Nvidia, Stanford, UC Berkeley, and UC San Diego introduced TTT-E2E, a method that compresses long context into transformer weights by training the model during inference via meta-learning. The approach uses sliding-window attention restricted to 8,000 tokens and updates only the fully connected layers of the last quarter of the network on each 1,000-token chunk at inference time, keeping per-token generation latency roughly constant as context scales to 128,000 tokens. TTT-E2E slightly outperforms vanilla transformers on next-token prediction loss across long contexts and matches efficient architectures like Mamba 2 and Gated DeltaNet on inference speed, but fails dramatically on Needle-in-a-Haystack retrieval beyond 8,000 tokens and incurs substantially higher training latency. The work reframes long-context handling as a training-inference trade-off rather than an architectural design problem.
Data Points: Qwen3.7-Max, OpenAI Math Proof, Gated DeltaNet-2, Trump AI Order, Microsoft Fara1.5
This edition of The Batch covers five significant AI developments: Alibaba's Qwen3.7-Max reasoning model with 1M token context and agentic capabilities ranking fifth on the Artificial Analysis Intelligence Index; an OpenAI reasoning model resolving the 80-year-old Erdős planar unit distance problem; Nvidia's Gated DeltaNet-2 outperforming Mamba-3 and other linear attention architectures; Trump pulling back a proposed AI regulation executive order; and Microsoft Research's Fara1.5 computer-use agent family beating OpenAI Operator and Google Gemini on the Online-Mind2Web benchmark.
MAST: Mechanism-guided selective unlearning for RLVR-trained reasoning models
Researchers introduce MAST (Mechanism-Aligned Selective Targeting), a method for selectively unlearning capabilities induced by reinforcement learning from verifiable rewards (RLVR) in language models while minimizing collateral damage to retained knowledge. The approach ranks attention-projection tensors by off-principal energy and gradient coupling to identify a targeted subset for update, rather than applying full-parameter gradient ascent. Evaluated on Qwen2.5-Math-1.5B and Qwen3-1.7B-Base, MAST achieves statistically significant forgetting on target MATH problems while preserving GSM8K performance, whereas full-parameter unlearning collapses retained capabilities. The method generalizes across seeds and unlearning objectives (NPO/SimNPO).
ADAS: Attention-Discounted Adaptive Sampler improves parallel decoding for masked diffusion language models
Researchers propose ADAS, a training-free reranking rule for masked diffusion language model decoding that addresses token interaction failures in parallel token commitment. The method greedily penalizes candidates that attend strongly to already-selected uncertain positions, using attention weights as soft marginal penalties rather than hard constraints. Evaluated on LLaDA-8B-Base and Dream-7B-Base across GSM8K, MATH500, HumanEval, and MBPP, ADAS improves low-NFE performance by 9–10 percentage points on average when plugged into existing samplers with only 3.1% runtime overhead.
DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention for Long-Context LLMs
DashAttention introduces a two-stage hierarchical sparse attention mechanism that replaces the fixed top-k block selection used in methods like NSA and InfLLMv2 with an adaptive α-entmax transformation, allowing a variable number of KV blocks to be selected per query. The approach keeps the full hierarchy differentiable by using the first-stage selection as a prior for second-stage softmax attention. Experiments show comparable accuracy to full attention at 75% sparsity with a better Pareto frontier than competing methods, and a Triton GPU implementation achieves meaningful speedup over FlashAttention-3 at inference time.
Neuronal Stochastic Attention Circuit (NSAC) for Probabilistic Representation Learning
Researchers introduce NSAC, a biologically-inspired continuous-time attention architecture that models attention logits as solutions to an Ornstein-Uhlenbeck stochastic differential equation, drawing on C. elegans Neuronal Circuit Policy wiring to induce Gaussian distributions over attention weights. The architecture enables joint quantification of aleatoric and epistemic uncertainty via a two-term objective combining Gaussian negative log-likelihood with an epistemic-separation regularizer. Empirical evaluation spans irregular time-series function approximation, multivariate regression, long-range forecasting, Industry 4.0 tasks, and autonomous vehicle lane-keeping, showing competitive accuracy with well-calibrated uncertainty estimates.