GPT-4.1 and o4-mini can generate correct plain BLAS C code for most of 20 routines from routine names alone, but optimized generations fail often and performance claims rest on best-of-10 selection with no error bars.
Attention is all you need,
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
-
Performance Evaluation of General Purpose Large Language Models for Basic Linear Algebra Subprograms Code Generation
GPT-4.1 and o4-mini can generate correct plain BLAS C code for most of 20 routines from routine names alone, but optimized generations fail often and performance claims rest on best-of-10 selection with no error bars.