COGNATE fine-tunes CPU-pretrained cost models on 500 accelerator samples, reaching 95% of optimal SpMM speedup on the SPADE simulator.
A cost model for compilers based on transfer learning
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
-
COGNATE: Acceleration of Sparse Tensor Programs on Emerging Hardware using Transfer Learning
COGNATE fine-tunes CPU-pretrained cost models on 500 accelerator samples, reaching 95% of optimal SpMM speedup on the SPADE simulator.