Hawkeye: Reproducing GPU-Level Non-Determinism
This addresses the need for efficient and trustworthy third-party auditing of ML model training and inference, which is an incremental improvement over prior verifiable machine learning approaches.
The authors tackled the problem of non-determinism in GPU arithmetic operations by developing Hawkeye, a system that enables exact reproduction of matrix multiplication on a CPU without precision loss, achieving perfect reproduction across multiple GPU architectures and precision types.
We present Hawkeye, a system for analyzing and reproducing GPU-level arithmetic operations. Using our framework, anyone can re-execute on a CPU the exact matrix multiplication operations underlying a machine learning model training or inference workflow that was executed on an NVIDIA GPU, without any precision loss. This is in stark contrast to prior approaches to verifiable machine learning, which either introduce significant computation overhead to the original model owner, or suffer from non-robustness and quality degradation. The main technical contribution of Hawkeye is a systematic sequence of carefully crafted tests that study rounding direction, subnormal number handling, and order of (non-associative) accumulation during matrix multiplication on NVIDIA's Tensor Cores. We test and evaluate our framework on multiple NVIDIA GPU architectures ( Ampere, Hopper, and Lovelace) and precision types (FP16, BFP16, FP8). In all test cases, Hawkeye enables perfect reproduction of matrix multiplication on a CPU, paving the way for efficient and trustworthy third-party auditing of ML model training and inference.