TESTING METHOD AND VERIFICATION PLATFORM FOR ACCELERATOR

摘要

A testing method and a verification platform for an accelerator. The testing method comprises: generating at least one target neural network according to a network description file (S1); translating the at least one target neural network into a neural network instruction (S2); respectively inputting the neural network instruction into an accelerator and a software model matching the accelerator for execution, and determining a difference between output results of the neural network instruction (S3); and according to the difference between the output results of the neural network instruction, determining an instruction that is anomalous in an accelerator running process (S4). According to the method, a plurality of target neural networks having a definite structure can be generated by generating at least one target neural network according to a network description file, effectively performing targeted performance testing on accelerators.