YunwenTechnology / Unilm

437 stars 86 forks source link

modeling_unilm部分的报错 #23

Open cuiMidAutumn opened 3 years ago

cuiMidAutumn commented 3 years ago

尊敬的作者,您好,在用您代码实验中modeling_unilm.py出现下面的报错, File "C:\Users\river\Unilm-master\modeling_unilm.py", line 431, in forward return self.beam_search(input_ids, token_type_ids, position_ids, attention_mask) File "C:\Users\river\Unilm-master\modeling_unilm.py", line 691, in beam_search seq.append(wids_list[fid - 1][pos_in_frame]) TypeError: list indices must be integers or slices, not float 在modeling_unilm.py报错的地方做如下的修改后, seq = [wids_list[frame_id][pos_in_frame]] for fid in range(frame_id, 0, -1): pos_in_frame = int(ptrs[fid][pos_in_frame]) # 添加了int() 类型转换 seq.append(wids_list[fid - 1][pos_in_frame]) seq.reverse() traces['pred_seq'].append(seq) 程序可正常运行,但出现预测输出文件重复字的情况。请问这样修改的对吗?重复字的问题是这样修改导致的吗?求指导,不甚感谢

Haruka1307 commented 1 year ago

我也有同样的问题