# ReplaySSM: Rethinking State Storage for Faster Hybrid-Model Decoding

**URL:** <https://forums.developer.nvidia.com/t/replayssm-rethinking-state-storage-for-faster-hybrid-model-decoding/385705>\
**Category:** Technical Blog\
**Tags:** nemotron\
**Created:** [October 12, 2026, 4:00pm UTC](https://forums.developer.nvidia.com/t/replayssm-rethinking-state-storage-for-faster-hybrid-model-decoding/385705 "2026-10-12T16:00:56Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![jwitsoe](https://sea2.discourse-cdn.com/nvidia/user_avatar/forums.developer.nvidia.com/jwitsoe/32/16591_2.png) [@jwitsoe](https://forums.developer.nvidia.com/u/jwitsoe)\
**Post date:** [October 12, 2026, 4:00pm UTC](https://forums.developer.nvidia.com/t/replayssm-rethinking-state-storage-for-faster-hybrid-model-decoding/385705/1 "2026-10-12T16:00:57Z")

</div>

Originally published at: [ReplaySSM: Rethinking State Storage for Faster Hybrid-Model Decoding | NVIDIA Technical Blog](https://developer.nvidia.com/blog/replayssm-rethinking-state-storage-for-faster-hybrid-model-decoding/)  
   
Hybrid models interleave state-space model (SSM) layers with a few attention layers. NVIDIA Nemotron 3, Qwen3.5 through Qwen3.8, Kimi-K3, and GLM-5.3-Flash are all built this way, with Mamba-2, Gated DeltaNet (GDN), or Kimi Delta Attention (KDA) as the SSM layer. SSM layers are designed to make decoding fast: they use a fixed-size state in place…