数值稳定的 Softmax 函数 通过减去最大值来避免数值溢出问题。 Args: logits: 未归一化的对数概率 [batch_size, num_classes] Returns: softmax: 归一化的概率分布 [batch_size, num_classes]
(logits)
source not stored for this graph (policy: none)
no outgoing calls
no test coverage detected