该量化后的模型可以在cpu上完成推理，在MLU上推理量化模型时报错，报错源码是在后处理锚框时postprocess()出现，源码如下，x为CV2读取照片的值，错误为索引错误，

def postprocess(x, anchors, regression, classification, regressBoxes, clipBoxes, threshold, iou_threshold):
    transformed_anchors = regressBoxes(anchors, regression)
    transformed_anchors = clipBoxes(transformed_anchors, x)
    scores = torch.max(classification, dim=2, keepdim=True)[0]
    scores_over_thresh = (scores > threshold)[:, :, 0]
    out = []
    for i in range(x.shape[0]):
        if scores_over_thresh[i].sum() == 0:
            out.append({
                'rois': np.array(()),
                'class_ids': np.array(()),
                'scores': np.array(()),
            })
            continue
        #a = torch.zeros(49104)
     

        #scores_over_thresh = scores_over_thresh(dtype=torch.long)  #socres_over_thresh[i, :]
        classification_per = classification[i, scores_over_thresh[i, :], ...].permute(1, 0)  #出错
        transformed_anchors_per = transformed_anchors[i, scores_over_thresh[i, :], ...]
        scores_per = scores[i, scores_over_thresh[i, :], ...]
        scores_, classes_ = classification_per.max(dim=0)
        anchors_nms_idx = batched_nms(transformed_anchors_per, scores_per[:, 0], classes_, iou_threshold=iou_threshold)

调试时发现scores_over_thresh[i, :]的值为49104个零，是float类型，尝试转换类型也不起作用，而且mlu上也不支持long类型，请问这个错该如何解决呢，如何把它变为mlu支持的正确的索引呢？