TransXSSM applies Rotary Position Embedding to both attention and state-space layers and reports speedups and accuracy gains over Transformer, Mamba, and Jamba baselines at 320M and 1.3B scale.
Variational learning for switching state-space models,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
TransXSSM: A Hybrid Transformer State Space Model with Unified Rotary Position Embedding
TransXSSM applies Rotary Position Embedding to both attention and state-space layers and reports speedups and accuracy gains over Transformer, Mamba, and Jamba baselines at 320M and 1.3B scale.