DF-SSM distills Mamba-2 to 1-bit scaffold plus int8 low-rank correction for 9.7x compression and 21.4x faster inference, plus analysis showing three distinct processing phases across layers.
The hidden attention of mamba models
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 2
citation-polarity summary
roles
background 2polarities
background 2representative citing papers
Transformers and SSMs are unified through structured state space duality, producing a 2-8X faster Mamba-2 model that remains competitive with Transformers.
Jamba presents a hybrid Transformer-Mamba MoE architecture for LLMs that delivers state-of-the-art benchmark performance and strong results up to 256K token contexts while fitting in one 80GB GPU with high throughput.
Content-based routing with a norm-blind fixed similarity metric produces representation collapse and concentration across attention, graph attention, state-space models, recurrent mixers, and residual connections.
citing papers explorer
-
Density Field State Space Models: 1-Bit Distillation, Efficient Inference, and Knowledge Organization in Mamba-2
DF-SSM distills Mamba-2 to 1-bit scaffold plus int8 low-rank correction for 9.7x compression and 21.4x faster inference, plus analysis showing three distinct processing phases across layers.
-
Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality
Transformers and SSMs are unified through structured state space duality, producing a 2-8X faster Mamba-2 model that remains competitive with Transformers.
-
Jamba: A Hybrid Transformer-Mamba Language Model
Jamba presents a hybrid Transformer-Mamba MoE architecture for LLMs that delivers state-of-the-art benchmark performance and strong results up to 256K token contexts while fitting in one 80GB GPU with high throughput.
-
All Routes Lead to Collapse
Content-based routing with a norm-blind fixed similarity metric produces representation collapse and concentration across attention, graph attention, state-space models, recurrent mixers, and residual connections.
- Advancing Intelligent Sequence Modeling: Evolution, Trade-offs, and Applications of State-Space Architectures from S4 to Mamba