A tile-level GPU language with an algebraic layout system generates kernels for arbitrary 1-8 bit quantized types, outperforming Triton, Ladder, QuantLLM, and Marlin on supported workloads.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Tilus: A Tile-Level GPGPU Programming Language for Low-Precision Computation
A tile-level GPU language with an algebraic layout system generates kernels for arbitrary 1-8 bit quantized types, outperforming Triton, Ladder, QuantLLM, and Marlin on supported workloads.