VibeServe demonstrates that AI agents can synthesize bespoke LLM serving systems end-to-end, remaining competitive with vLLM in standard settings while outperforming it in six non-standard scenarios involving unusual models, workloads, or hardware.
hub Canonical reference
Nemotron-H: A family of accurate and efficient hybrid Mamba-Transformer models
Canonical reference. 86% of citing Pith papers cite this work as background.
22
Pith papers citing it
Background
86% of classified citations
abstract
As inference-time scaling becomes critical for enhanced reasoning capabilities, it is increasingly becoming important to build models that are efficient to infer. We introduce Nemotron-H, a family of 8B and 56B/47B hybrid Mamba-Transformer models designed to reduce inference cost for a given accuracy level. To achieve this goal, we replace the majority of self-attention layers in the common Transformer model architecture with Mamba layers that perform constant computation and require constant memory per generated token. We show that Nemotron-H models offer either better or on-par accuracy compared to other similarly-sized state-of-the-art open-sourced Transformer models (e.g., Qwen-2.5-7B/72B and Llama-3.1-8B/70B), while being up to 3$\times$ faster at inference. To further increase inference speed and reduce the memory required at inference time, we created Nemotron-H-47B-Base from the 56B model using a new compression via pruning and distillation technique called MiniPuzzle. Nemotron-H-47B-Base achieves similar accuracy to the 56B model, but is 20% faster to infer. In addition, we introduce an FP8-based training recipe and show that it can achieve on par results with BF16-based training. This recipe is used to train the 56B model. We are releasing Nemotron-H base model checkpoints with support in Hugging Face and NeMo.
hub tools
citation-role summary
background 6
baseline 1
citation-polarity summary
representative citing papers
FlashMorph formulates hybrid layer selection as budget-constrained optimization, trains per-layer gates on synthetic retrieval data with linearization regularization, then discretizes and distills to produce efficient hybrid architectures.
MOSAIC is a simulation and DSE framework for heterogeneous NPUs that finds designs achieving 46.91% mean iso-area energy savings over homogeneous baselines on 20 workloads.
SISA adds an SSM importance term inside the attention score and runs the full operation as one SDPA call on augmented Q/K vectors, reporting better LAMBADA and perfect NIAH at small scale.
HEED replaces uniform residual alignment with density-weighted alignment using patch self-dissimilarity to improve hybrid VLM distillation, gaining 8.7 points on OCRBench v2 and 5.13 on a 10-benchmark average.
Star Elastic trains N nested submodels in a single post-training job on a parent reasoning LLM, supporting elastic budget control that matches or exceeds independent baselines while cutting training compute by up to 360x.
Short input phrases can irreversibly overwrite hidden states in Mamba models, impairing information retrieval on a new benchmark while leaving pure Transformer models unaffected.
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
Reinforcement learning after SFT conversion narrows the performance gap between sliding-window attention and full self-attention on math reasoning benchmarks while preserving linear complexity.
Flash PD-SSM achieves FSA-level expressivity by discretely selecting one matrix from a trainable set of structured sparse transition matrices at each time step while preserving the runtime and memory efficiency of standard structured SSMs.
DashAttention introduces differentiable adaptive sparse hierarchical attention via α-entmax block selection, achieving full-attention accuracy at 75% sparsity with improved Pareto performance over NSA and InfLLMv2.
Attention decomposes into low-rank routing and symmetric filtering; disentangled S-D attention reveals a spectral cascade allowing early-layer linearization at under 5% perplexity cost.
ModelLens learns a performance-aware latent space from 1.62M leaderboard records to rank unseen models on unseen datasets without forward passes on the target.
No model can achieve efficiency, compactness, and recall capacity scaling with sequence length at once, as any two imply a strict bound of O(poly(d)/log V) on recallable facts.
ABMamba uses Mamba-based linear-complexity processing plus a novel Aligned Hierarchical Bidirectional Scan to deliver competitive video captioning on VATEX and MSR-VTT at roughly 3x higher throughput than typical Transformer MLLMs.
Kimi Linear hybridizes linear attention with a new KDA module to beat full attention on tasks while slashing KV cache by 75% and speeding decoding up to 6x.
Short sliding windows in hybrid attention-xLSTM models boost long-context performance by encouraging long-term memory use, and stochastic window sizing improves both short and long tasks.
Cosmos-Reason1-7B and 56B models are trained with physical common sense and embodied reasoning ontologies via supervised fine-tuning and reinforcement learning to produce next-step physical actions.
MOSAIC uses an Integer Linear Program scheduler for expert placement and prompt assignment plus adaptive aggregation to achieve 1.7-2.3x end-to-end speedup on 4-GPU MoA workloads while keeping accuracy within 0.1pp.
Mambalaya fuses the entire Mamba layer into one on-chip computation group, achieving simulated 4.9x prefill and 1.9x generation speedups over a MARCA-like baseline.
NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.
Small language models are sufficiently capable, more suitable, and far more economical than large models for the repetitive tasks that dominate agentic AI systems.
citing papers explorer
-
VibeServe: Can AI Agents Build Bespoke LLM Serving Systems?
VibeServe demonstrates that AI agents can synthesize bespoke LLM serving systems end-to-end, remaining competitive with vLLM in standard settings while outperforming it in six non-standard scenarios involving unusual models, workloads, or hardware.
-
Morphing into Hybrid Attention Models
FlashMorph formulates hybrid layer selection as budget-constrained optimization, trains per-layer gates on synthetic retrieval data with linearization regularization, then discretizes and distills to produce efficient hybrid architectures.
-
MOSAIC: A Workload-Driven Simulation and Design-Space Exploration Framework for Heterogeneous NPUs
MOSAIC is a simulation and DSE framework for heterogeneous NPUs that finds designs achieving 46.91% mean iso-area energy savings over homogeneous baselines on 20 workloads.
-
Forget Attention: Importance-Aware Attention Is All You Need
SISA adds an SSM importance term inside the attention score and runs the full operation as one SDPA call on augmented Q/K vectors, reporting better LAMBADA and perfect NIAH at small scale.
-
HEED: Density-Weighted Residual Alignment for Hybrid Vision-Language Model Distillation
HEED replaces uniform residual alignment with density-weighted alignment using patch self-dissimilarity to improve hybrid VLM distillation, gaining 8.7 points on OCRBench v2 and 5.13 on a 10-benchmark average.
-
Star Elastic: Many-in-One Reasoning LLMs with Efficient Budget Control
Star Elastic trains N nested submodels in a single post-training job on a parent reasoning LLM, supporting elastic budget control that matches or exceeds independent baselines while cutting training compute by up to 360x.
-
Hidden State Poisoning Attacks against Mamba-based Language Models
Short input phrases can irreversibly overwrite hidden states in Mamba models, impairing information retrieval on a new benchmark while leaving pure Transformer models unaffected.
-
Unified Audio Intelligence Without Regressing on Text Intelligence
Audex unifies audio understanding and generation on a strong text MoE backbone with multi-stage SFT plus text-only Cascade RL, matching open SOTA audio scores while mostly retaining text capability.
-
Architecture-Aware Reinforcement Learning Makes Sliding-Window Attention Competitive in Math Reasoning
Reinforcement learning after SFT conversion narrows the performance gap between sliding-window attention and full self-attention on math reasoning benchmarks while preserving linear complexity.
-
Flash PD-SSM: Memory-Optimized Structured Sparse State-Space Models
Flash PD-SSM achieves FSA-level expressivity by discretely selecting one matrix from a trainable set of structured sparse transition matrices at each time step while preserving the runtime and memory efficiency of standard structured SSMs.
-
DashAttention: Differentiable and Adaptive Sparse Hierarchical Attention
DashAttention introduces differentiable adaptive sparse hierarchical attention via α-entmax block selection, achieving full-attention accuracy at 75% sparsity with improved Pareto performance over NSA and InfLLMv2.
-
The Routing and Filtering Structure of Attention
Attention decomposes into low-rank routing and symmetric filtering; disentangled S-D attention reveals a spectral cascade allowing early-layer linearization at under 5% perplexity cost.
-
ModelLens: Finding the Best for Your Task from Myriads of Models
ModelLens learns a performance-aware latent space from 1.62M leaderboard records to rank unseen models on unseen datasets without forward passes on the target.
-
The Impossibility Triangle of Long-Context Modeling
No model can achieve efficiency, compactness, and recall capacity scaling with sequence length at once, as any two imply a strict bound of O(poly(d)/log V) on recallable facts.
-
ABMAMBA: Multimodal Large Language Model with Aligned Hierarchical Bidirectional Scan for Efficient Video Captioning
ABMamba uses Mamba-based linear-complexity processing plus a novel Aligned Hierarchical Bidirectional Scan to deliver competitive video captioning on VATEX and MSR-VTT at roughly 3x higher throughput than typical Transformer MLLMs.
-
Kimi Linear: An Expressive, Efficient Attention Architecture
Kimi Linear hybridizes linear attention with a new KDA module to beat full attention on tasks while slashing KV cache by 75% and speeding decoding up to 6x.
-
Short window attention enables long-term memorization
Short sliding windows in hybrid attention-xLSTM models boost long-context performance by encouraging long-term memory use, and stochastic window sizing improves both short and long tasks.
-
Cosmos-Reason1: From Physical Common Sense To Embodied Reasoning
Cosmos-Reason1-7B and 56B models are trained with physical common sense and embodied reasoning ontologies via supervised fine-tuning and reinforcement learning to produce next-step physical actions.
-
MOSAIC: Efficient Mixture-of-Agent Scheduling via Adaptive Aggregation and Inference Concurrency
MOSAIC uses an Integer Linear Program scheduler for expert placement and prompt assignment plus adaptive aggregation to achieve 1.7-2.3x end-to-end speedup on 4-GPU MoA workloads while keeping accuracy within 0.1pp.
-
Mambalaya: Einsum-Based Fusion Optimizations on State-Space Models
Mambalaya fuses the entire Mamba layer into one on-chip computation group, achieving simulated 4.9x prefill and 1.9x generation speedups over a MARCA-like baseline.
-
NVIDIA Nemotron 3: Efficient and Open Intelligence
NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.
-
Small Language Models are the Future of Agentic AI
Small language models are sufficiently capable, more suitable, and far more economical than large models for the repetitive tasks that dominate agentic AI systems.