Systematic study of direct and iterative triangular matrix inversion algorithms for Delta-Rule linear transformers, with focus on numerical stability, hardware efficiency, and benchmarks showing 4.3x NPU speedup over SGLang.
Condition numbers of random triangular matrices.SIAM Jounrnal on Matrix Analysis and Applications, 19(2):564–581
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Fast and Stable Triangular Inversion for Delta-Rule Linear Transformers
Systematic study of direct and iterative triangular matrix inversion algorithms for Delta-Rule linear transformers, with focus on numerical stability, hardware efficiency, and benchmarks showing 4.3x NPU speedup over SGLang.