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.
Related guides (3)
Related events (8)
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.
Bamba: Inference-Efficient Hybrid Mamba2 Model
Hugging Face published a blog post introducing Bamba, a hybrid architecture combining Mamba2 state-space layers with attention layers, designed for inference efficiency. The model targets reduced KV-cache memory and improved throughput compared to pure transformer architectures. The post covers architecture details, training approach, and benchmarking results positioning Bamba as a practical alternative for deployment-constrained settings.
Trajectory Analysis of Masked Diffusion LMs for Graph-to-Text Generation with Lambda-Scaled Structural Decoding
This paper presents the first systematic study of masked diffusion language models (MDLMs) for graph-to-text generation, analyzing the order in which tokens are unmasked during iterative decoding. The authors find MDLMs naturally unmask entities first, then relational/function words, then structural tokens—a pattern disrupted by supervised fine-tuning, which prematurely anchors structural tokens and causes hallucination or omission. They propose lambda-scaled structural decoding, a training-free inference-time fix that recovers +9.4 BLEU-4, and introduce Graph-LLaDA, which integrates a Graph Transformer encoder into LLaDA's decoding process. Cross-dataset evaluation on the LAGRANGE benchmark shows prior baselines overfit to dataset-specific patterns while MDLM-based approaches generalize better.
Sleep paradigm for LLMs enables continual learning and memory consolidation via distillation and RL
A new arXiv preprint proposes a 'Sleep' paradigm for language models that enables continual learning by consolidating short-term in-context memories into long-term parameters. The framework has two stages: Knowledge Seeding (distilling a smaller model's memories into a larger network via on-policy distillation combined with RL-based imitation learning) and Dreaming (self-improvement via RL-generated synthetic curricula without human supervision). Experiments cover long-horizon tasks, continual learning, knowledge incorporation, and few-shot generalization, addressing a known weakness of current LLMs in retaining temporal knowledge across contexts.
Codestral Mamba: Mistral AI Releases Apache 2.0 Mamba-Architecture Code Model
Mistral AI has released Codestral Mamba, a 7.3B-parameter code-focused language model built on the Mamba state-space architecture rather than the Transformer architecture. The model offers linear-time inference and theoretically infinite sequence length, tested up to 256k tokens in-context retrieval. Developed with Mamba co-creators Albert Gu and Tri Dao, it is released under Apache 2.0 and available via HuggingFace, mistral-inference SDK, TensorRT-LLM, and Mistral's la Plateforme API. Mistral positions it as a local code assistant that performs on par with state-of-the-art transformer-based code models.
Dynamic short convolutions yield 1.33–1.60× compute advantage over standard Transformers
A new arXiv preprint introduces dynamic short convolutions as an architectural primitive for Transformers, using input-dependent filters to combine locality bias with increased expressivity. Experiments across 150M–2B parameter language models show consistent perplexity improvements over standard Transformers and static convolution variants, with scaling-law fits indicating a 1.33× compute advantage when applied to key/query/value vectors and 1.60× when added after every linear layer. The technique also improves linear RNNs (Mamba-2, Gated DeltaNet) and mixture-of-experts architectures, with custom Triton kernels making training practical.
Comparing RoBERTa, Llama 2, and Mistral for Sequence Classification via LoRA on Disaster Tweets
A Hugging Face blog post benchmarks three models—RoBERTa, Llama 2, and Mistral—on a disaster tweet classification task using LoRA fine-tuning. The analysis compares parameter-efficient adaptation of encoder-only versus decoder-only architectures for a practical NLP classification problem. Results provide practitioners with guidance on model selection and LoRA configuration for sequence classification.
Falcon Mamba: First Strong Attention-Free 7B Model
Technology Innovation Institute (TII) releases Falcon Mamba, a 7B parameter state space model (SSM) based on the Mamba architecture, announced as the first attention-free model at this scale to match or exceed transformer-based models on standard benchmarks. The model is hosted on Hugging Face and represents a significant milestone for SSM-based architectures competing with transformers. This release advances the case for pure SSM models as viable alternatives to attention-based LLMs at the 7B scale.