MixT compresses Transformer LLMs by substituting targeted linear projections with tensor-operator mixtures, preserving MMLU accuracy up to model-specific boundaries where parameter count drops 47.5% and inference memory 60.4% on LLaMA2-7B.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
A general tensor-structured compression scheme for efficient large language models
MixT compresses Transformer LLMs by substituting targeted linear projections with tensor-operator mixtures, preserving MMLU accuracy up to model-specific boundaries where parameter count drops 47.5% and inference memory 60.4% on LLaMA2-7B.