Beyond atol and rtol: Sound Test Oracles for Low-Precision GPU Kernels
cs.MS, cs.PL, cs.SE
Submitted: 2026-10-28
Updated: 2026-11-09
Comments: 37 pages, 5 figures
Code: https://github.com/ROCm/aiter
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Automated Numerical Stability Analysis of Deep Learning Operators
- DeepSeek-V3 Technical Report
- Measuring the Checker: Mutation Analysis for GPU-Kernel Benchmark Oracles
- Mixed precision accumulation for neural network inference guided by componentwise forward error analysis
- V-ABFT: Variance-Based Adaptive Threshold for Fault-Tolerant Matrix Multiplication in Mixed-Precision Deep Learning
- Liger Kernel: Efficient Triton Kernels for LLM Training
- FP8 Formats for Deep Learning
- Microscaling Data Formats for Deep Learning
- The Correctness Illusion in LLM-Generated GPU Kernels
- Operator-Aware Mixed-Precision Tolerance Calibration for Tensor Kernels
- A Contract-Grade Verifier for LLM-Generated GPU Kernels, and a Native Blackwell Backward for the Gated-Linear-Recurrence Family
- Kernel Contracts: A Specification Language for ML Kernel Correctness Across Heterogeneous Silicon
Related papers
- asdex: Automatic Sparse Differentiation in JAX
- Lord of the Precision, the Two Floats: Double-Word Arithmetic for Complex Numbers
- Streaming Right Multiplication over Grammar-Compressed Matrices: A Memory-Bounded GPU Engine for Genotype and Graph Data
- Improved Scaling for Fast Mode of Ozaki Scheme II
- BF16 Component-Product Emulation of FP32 and FP64 GEMM on Intel AMX
- QRecoupling.jl: numerical and exact computation of classical and quantum recoupling coefficients