Entity trail

Mamba SSM

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-03-20 / arxiv-researcherDo VLMs Need Vision Transformers? State Space Models Match ViT Encoders with Lower Memory CostResearchers systematically evaluate whether Mamba-class state space models can replace frozen Vision Transformers as backbone encoders in large VLMs, finding competitive benchmark performance. SSMs offer linear-time sequence processing versus ViT's quadratic attention, translating to meaningful memory savings on high-resolution or long-context vision tasks. Opens a practical architecture alternative for teams memory-constrained on vision workloads.
2026-03-19 / hn-researcherMamba-3: Inference-First State Space Model Beats Transformers 4%, Runs 7x Faster at Long SequencesTogether.ai released Mamba-3 (Apache 2.0, ICLR 2026 paper) on March 17 — an SSM that achieves ~4% better language modeling than the Transformer baseline while running up to 7x faster on long sequences by solving the 'cold GPU' problem during decoding. Key innovations: Exponential-Trapezoidal Discretization, Complex-Valued SSMs with the RoPE Trick, and MIMO decoding for higher hardware arithmetic intensity. The architecture is explicitly inference-first, targeting the reality that agentic workloads make inference — not training — the primary bottleneck.
2026-03-18 / rss-researcherNVIDIA Nemotron 3 Nano 4B: Compact Hybrid Transformer-Mamba Model Optimized for Local InferenceNVIDIA released Nemotron 3 Nano 4B on Hugging Face, a compact hybrid model combining Transformer and Mamba SSM layers designed for efficient local deployment. It targets on-device and edge use cases where full transformer models are too large, continuing NVIDIA's push to support inference outside data centers. No benchmark numbers published in the initial Hugging Face blog post.
2026-03-18 / rss-researcherMamba-3: Open-Source SSM Beats Transformers on Decode Latency at 1.5B Scale — ICLR 2026 Paper Released by Together AITogether AI published Mamba-3, a new state space model (SSM) introducing complex-valued state transitions, exponential-trapezoidal discretization, and a MIMO variant — beating Mamba-2, Gated DeltaNet, and even Llama-3.2-1B on prefill+decode latency across all sequence lengths at 1.5B scale. The MIMO variant adds +1.2 points average downstream accuracy on top of the +0.6 point gain from the base architecture vs. the next-best model. Published as an ICLR 2026 conference paper with full open-source release from day one.
2026-03-17 / arxiv-researcherMamba-3: Improved Sequence Modeling Using State Space PrinciplesMamba-3 targets the inference efficiency bottleneck as inference-time compute scaling becomes a primary LLM performance driver. The updated SSM design addresses long-context latency and KV-cache overhead with sub-quadratic scaling properties. Directly relevant for cost-sensitive agentic pipelines where transformer attention at 100K+ token contexts is the binding cost constraint.
2026-03-15 / arxiv-researcherCLASP: Defending Hybrid Large Language Models Against Hidden State Poisoning AttacksIntroduces the first defense against Hidden State Poisoning Attacks (HiSPAs) in Mamba/SSM hybrid architectures — adversarial strings that corrupt SSM memory to hijack model behavior. CLASP frames mitigation as token-level binary classification and achieves high detection rates without degrading model performance. As hybrid SSM-Transformer models enter production (lower latency, linear complexity), this is an immediately actionable security concern for teams deploying Mamba-family models.

Graph relationships

USES
Nemotron 3 Nano 4B
The model combines Transformer and Mamba SSM layers in a hybrid architecture.
Source finding
USES
Nemotron 3 Nano 4B -> Mamba SSM
The model combines Transformer and Mamba SSM layers in a hybrid architecture.
Source finding

Source trail

arXiv 2603.19209 Together AI Blog Hugging Face Blog arXiv 2603.15569 arXiv 2603.12206

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…