attention_scores = inputs @ inputs.T
attention_weights = torch.softmax(attention_scores, dim=-1)
context_vectors = attention_weights @ inputs

        __
        __