RAD selects redundant attention layers via self-speculative decoding throughput, replaces them with SSM blocks, and distills them into hybrid models that beat the original on GSM8K and CRUX.
Xing, J Zico Kolter, and Albert Gu
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
background 1representative citing papers
citing papers explorer
-
RAD: Redundancy-Aware Distillation for Hybrid Models via Self-Speculative Decoding
RAD selects redundant attention layers via self-speculative decoding throughput, replaces them with SSM blocks, and distills them into hybrid models that beat the original on GSM8K and CRUX.