XiaokunDing

Results 1 comments of XiaokunDing

When exporting a model from PyTorch to ONNX using float16 precision, there is a significant difference in the output of the following operator. ` // Attention heads [sq, b, h]...