Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

运行错误,求教 #5

Open
YasinZhao opened this issue Apr 11, 2019 · 8 comments
Open

运行错误,求教 #5

YasinZhao opened this issue Apr 11, 2019 · 8 comments

Comments

@YasinZhao
Copy link

你好,在迭代一次完成进行评估时,发生如下错误,请问这是什么原因导致的呢,在--prepare阶段有--pretrain过
image

@SeekPoint
Copy link

me too

@shiningliang
Copy link
Owner

@YasinZhao @lovejasmine 感谢两位的反馈,近期我会检查一下

@wangbo3
Copy link

wangbo3 commented Aug 15, 2019

您好方便留个联系方式吗,以后多多学习

@fooSynaptic
Copy link

same issue

@fooSynaptic
Copy link

我好像发现了问题,这个prepare的程序处理的过程中,对于多个passage只保留了passage_token_ids却没有保留passage_tokens,因此对于上面那个问题可以用passage_token_ids来绕过求句子长度的问题,但是对于后面生成计算出来的结果文本时只有token的idx信息是不够的,这个问题应该可以通过检查prepare来解决

{'answer_spans': [[12, 25]], 'answers': ['1.《将夜》2.《择天记》3.《冒牌大英雄》4.《无限恐怖》5.《恐怖搞校》6.《大国医》7.《龙魔导》。', '《大唐悬疑录:长恨歌密码》、《风雪追击》、《草原动物园》、《有匪2:离恨楼》。', '我们住在一起、月都花落,沧海花开、天定风华、寻找爱情的邹小姐、应许之日、星光的彼端、他来了,请闭眼。'], 'passages': [{'passage_token_ids': [304, 9027, 5274, 10069, 11144, 2439, 7282, 5935, 9027, 8917, 10208, 10208, 16431, 14496, 1, 6204, 5918, 1694, 16706, 1694, 10640, 1, 8917, 7827, 12329, 4047, 11144, 13612, 412, 14213, 16432, 4615, 648, 5154, 8984, 13342, 15671, 6180, 8962, 3827, 2827, 5154, 1, 8917, 1, 11144, 6347, 268, 5154, 11732, 1963, 12245, 6883, 3577, 4439, 16726, 10019, 13587, 10700, 6204, 3346, 1, 1, 8917, 1, 6204, 6347, 11144, 14858, 5154, 8685, 412, 7075, 8116, 5154, 8984, 6204, 16667, 3474, 7901, 8917, 10153, 1, 6204, 14765, 11165, 2017, 6204, 5274, 11144, 321, 412, 1891, 8116, 5154, 8984, 11633, 8404, 6204, 7646, 9174, 1418, 8917, 16989, 15343, 2915, 8917, 12182, 4094, 4817, 3676, 9867, 9325, 11144, 10183, 412, 6223, 8116, 5154, 1, 9677, 1, 8917, 7271, 6347, 10019, 1963, 6204, 7438, 8504, 11144, 2881, 412, 2915, 12989, 10640, 9983, 1963, 13987, 10518, 8917, 7271, 8685, 9597, 9917, 4184, 10149, 6204, 8884, 11144, 13342, 12639, 119, 119, 1618, 7271, 15610, 17734, 6204, 9219, 11144, 3827, 11165, 16431, 14765, 1, 6204, 11058, 5154, 11453, 11144, 11453, 552, 5100, 8917, 887, 10713, 10019, 7094, 8917, 15830, 10550, 5100, 8917, 1, 266, 1, 11144, 13342, 14067, 10640, 11945, 4475, 11144, 3827, 11453, 552, 13690, 8917, 13690, 1, 13976, 1, 8917, 14329, 6642, 11469, 5843, 1, 7608, 15610, 16813, 16132, 8917, 2830, 21, 1963, 5908, 16584, 16980, 8917, 12402, 1489, 6510, 2806, 11144, 11453, 552, 195, 8917, 195, 2924, 1, 8917, 17119, 15610, 3346, 1, 17573, 2948, 6204, 1, 8917, 17119, 15449, 14319, 15839, 8613, 8161, 6204, 10030, 4477, 17331, 8695, 8161, 11669, 11144, 11453, 552, 10183, 1661, 8917, 10183, 1661, 1, 8917, 6972, 119, 119, 1, 10022, 7271, 8569, 10713, 8917, 4911, 12159, 1088, 16745, 12474, 8917, 8728, 14989, 1856, 9867, 4717, 14765, 16084, 1, 11144, 11129, 2732, 8917, 1618, 7271, 8685, 16706, 10121, 6204, 5274, 11144, 10020, 412, 14213, 15668, 4615, 648, 5154, 8984, 1, 13342, 12639, 119, 119, 1, 8917, 14496, 11203, 15449, 7743, 16223, 8984, 11144, 3827, 16903, 16432, 8917, 1618, 7271, 8685, 16706, 1761, 1088, 6204, 15078, 8917, 1, 11144, 453, 16099, 17226, 11695, 8917, 17333, 11732, 16337, 8917, 1651, 7271, 5097, 8917, 14496, 9410, 9418, 887, 5097, 6204, 3346, 298, 8917, 10421, 14496, 2917, 7709, 16337, 8632, 10729, 8980, 16267, 8917, 16401, 14202, 1856, 8917, 13835, 10309, 12188, 252, 8917, 17028, 3495, 2963, 17573, 12059, 12985, 11144, 1953, 412, 14213, 1, 4615, 648, 5154, 5454, 7282, 321, 3558, 11144, 12163, 8917, 5765, 13865, 6204, 3433, 10713, 8917, 13749, 1, 17214, 15397, 6204, 11144, 1088, 12163, 6204, 648, 8685, 16706, 1, 8917, 17638, 12163, 13217, 6204, 9174, 8116, 11144, 3646, 12163, 4359, 1, 11144, 887, 5765, 13865, 6204, 8917, 2539, 1, 9366, 12163, 8917, 1, 11144, 12738, 412, 14213, 1, 4615, 648, 5154, 1, 14679, 13865, 1, 7608, 9831, 6204, 3433, 10713, 11144, 887, 14679, 13865, 6204, 8917, 10121, 11144, 14476, 412, 14213, 11531, 7166, 9861, 4615, 648, 5154, 1, 11062, 2004, 3389, 12163, 14853, 14679, 13865, 9983, 6780, 8705, 8917, 9174, 1, 11144, 13470, 412, 14213, 1, 4615, 648, 5154, 5303, 1, 1618, 7271, 8685, 13074, 4283, 14782, 6204, 10713, 8917, 16223, 14496, 16561, 9366, 9091, 6204, 4283, 14782, 11144, 13835, 4589, 14213, 2186, 4615, 4468, 4283, 14782, 6204, 8917, 17316, 1618, 16706, 7271, 4677, 321, 16706, 11144, 5270, 11732, 13862, 6510, 1, 2460, 8685, 14046, 7516, 12028, 6204, 6347, 11144, 3612, 1, 8917, 3176, 11144, 8504, 9410, 8349, 8917, 3447, 3447, 11144, 16570, 412, 14213, 10795, 1, 4615, 648, 5154, 15449, 1, 6204, 1, 8467, 6044, 5154, 15397, 10372, 12182, 10223, 54, 1, 10372, 1544, 15686, 13082, 3263, 17403, 1, 6204, 9818, 11144, 8827, 4359, 1489, 6510, 1, 6204, 2529, 8642, 11144, 8827, 5154, 4114, 9811, 10166, 10871, 10372, 14573, 11662, 16431, 14496, 10733, 119, 119, 8014, 15397, 10372, 12182, 10223, 347, 6510, 10372, 1544, 15686, 13082, 3263, 17573, 16759, 14683, 11480, 12888, 16980, 11144, 8827, 21, 14535, 8917, 12405, 8753, 9665, 13659, 7714, 15276, 1, 8899, 1968, 119, 119, 7281, 4046, 5154, 4114, 7827, 17566, 16713, 8917, 14496, 9418, 16833, 6204, 8917, 4911, 1, 3346, 2696, 11144, 8014, 6347, 7185, 5154, 11732, 8827, 21, 930, 8259, 1, 4306, 6204, 2148, 10019, 6121, 6510, 8685, 10699, 1, 2148, 7443, 11144, 10421, 16548, 6636, 3298, 1618, 10699, 2148, 7443, 2830, 8917, 21, 2148, 10019, 16007, 6204, 9093, 2850, 604, 5998, 6883, 11610, 7323, 11144, 15453, 21, 11732, 6204, 8685, 1910, 4028, 11669, 8917, 12402, 8733, 14761, 5122, 6204, 1493, 3638, 2148, 8917, 13152, 17656, 11610, 7323, 11144, 8578, 8917, 11732, 6416, 6204, 9517, 4911, 13670, 1, 11144, 14858, 5154, 13612, 412, 7075, 7176, 8917, 9457, 3542, 9539, 8917, 552, 16432, 7608, 15331, 11144, 10020, 412, 6223, 7176, 8917, 11732, 7271, 8685, 9385, 6565, 1, 6204, 1, 11144, 13342, 1, 6204, 4805, 16881, 14496, 6684, 6510, 11144, 3827, 1953, 412, 1891, 7176, 8917, 2335, 8060, 2914, 8917, 10421, 6252, 8685, 10320, 1, 6204, 8941, 8917, 9141, 11144, 4267, 8828, 14573, 13342, 1, 3827]}, {'passage_token_ids': [1, 10069, 1, 10069, 1, 10069, 4850, 9666, 10069, 9729, 9666, 10069, 1, 10069, 2148, 9666, 10069, 872, 9666, 10069, 1, 10069, 13843, 9666, 10069, 1, 10069, 1, 10069, 1, 9666, 10069, 1, 10069, 10156, 9666, 10069, 14349, 9666, 10069, 1710, 9666, 10069, 1, 10069]}, {'passage_token_ids': [4073, 8917, 10156, 9666, 1650, 10156, 1856, 15610, 12059, 1650, 1, 6204, 17708, 4857, 17047, 845, 6204, 16342, 1544, 10019, 8917, 6328, 7403, 6510, 8152, 6204, 1, 8917, 8757, 8685, 7590, 12550, 6204, 4114, 10156, 8014, 7608, 15331, 11144, 14573, 6204, 10156, 9666, 10019, 8917, 11059, 6204, 12471, 10156, 9666, 552, 1177, 10372, 12025, 7271, 3346, 3558, 8828, 7968, 13083, 6204, 17347, 304, 2402, 11059, 6204, 10156, 9666, 8917, 7004, 5644, 1145, 12182, 887, 6204, 13788, 17758]}, {'passage_token_ids': [304, 9027, 6204, 4677, 8334, 13932, 11025, 8917, 13083, 21, 1618, 11482, 5833, 552, 11890, 17047, 15397, 5274, 2101, 10372, 1, 14329, 1743, 5154, 4114, 1, 16980, 8917, 17237, 5843, 1, 11144, 1, 16980, 8917, 6451, 3180, 15821, 7445, 11144, 1, 8884, 16980, 8917, 6451, 14832, 4128, 2169, 11144, 8014, 16989, 9777, 12182, 7271, 8685, 2903, 1, 21, 1, 10019, 1, 6204, 7625, 8917, 15276, 8685, 2903, 4359, 12164, 8349, 9027, 6204, 17770, 1, 8917, 2439, 5342, 16989, 955, 3387, 1, 6204, 8637, 8917, 13152, 955, 3577, 10019, 15343, 7323, 1, 6204, 3529, 1, 8917, 9410, 955, 3577, 10019, 1, 600, 9983, 1498, 16729, 11144]}, {'passage_token_ids': [17327, 6870, 14613, 8178, 12873, 648, 5154, 15622, 8685, 6661, 2827, 5154, 14211, 4276, 8917, 6870, 5935, 8115, 7271, 17327, 13891, 11444, 1650, 2402, 3213, 600, 6204, 13327, 11144, 6642, 12989, 10508, 8349, 4359, 17573, 1, 13976, 3031, 8917, 1, 6510, 1, 1650, 13123, 9983, 16563, 13897, 8917, 4292, 1, 600, 8917, 12989, 13304, 4635, 1, 11090, 6204, 16177, 8917, 16652, 16688, 10358, 14645, 6883, 11090, 6204, 8313, 11144, 9410, 14211, 4276, 17327, 6870, 8115, 13304, 8917, 4911, 5182, 6510, 1, 6204, 13327, 8917, 9345, 9832, 2239, 552, 357, 14357, 8917, 17098, 1478, 552, 6719, 14683, 10259, 6204, 9818]}], 'question_id': 186572, 'answer_passages': [3], 'match_scores': [0.43902439024390205], 'fact_or_opinion': 'OPINION', 'question_token_ids': [304, 552, 15397, 11059, 6204, 9666], 'entity_answers': [['将夜', '择天记', '冒牌大英雄', '无限恐怖', '恐怖搞校', '大国医', '龙魔导'], ['长恨歌密码', '风雪追击', '草原动物园', '离恨楼'], ['我们住在一起', '月都花落', '沧海花开', '天定风华', '寻找爱情的邹小姐', '应许之日', '星光的彼端', '他来了', '请闭眼']], 'segmented_question': ['2017', '有', '什么', '好看', '的', '小说'], 'question_type': 'ENTITY'}

@fooSynaptic
Copy link

完整的debug过程可以通过修改代码:

    def find_best_answer(self, sample, start_prob, end_prob, padded_p_len):
        """
        Finds the best answer for a sample given start_prob and end_prob for each position.
        This will call find_best_answer_for_passage because there are multiple passages in a sample
        """
        best_p_idx, best_span, best_score = None, None, 0
        for p_idx, passage in enumerate(sample['passages']):
            if p_idx >= self.max_p_num:
                continue
            print(passage.keys(), [len(passage[k]) for k in passage.keys()])
            #assert False
            passage_len = min(self.max_p_len, (len(passage['passage_tokens']) if 'passage_tokens' in passage else len(passage['passage_token_ids'])))
            # 为每个passage找到best answer
            answer_span, score = self.find_best_answer_for_passage(
                start_prob[p_idx * padded_p_len: (p_idx + 1) * padded_p_len],
                end_prob[p_idx * padded_p_len: (p_idx + 1) * padded_p_len],
                passage_len)
            # 各passage间最大score
            if score > best_score:
                best_score = score
                best_p_idx = p_idx
                best_span = answer_span
        # 根据span找到token
        if best_p_idx is None or best_span is None:
            best_answer = ''
        else:
            best_answer_idx = sample['passages'][best_p_idx]['passage_token_ids'][best_span[0]: best_span[1] + 1]
            best_answer = self.vocab.recover_from_ids(best_answer_idx)
            print(best_span, best_answer_idx, best_answer)

            #sample['passages'][best_p_idx]['passage_tokens'][best_span[0]: best_span[1] + 1])
        return best_answer

在我最后一轮迭代中get_best_answerprint的结果如下

2019-08-20 13:57:13,634 - brc - INFO - Saving dev.predicted results to ../data/results/dev.predicted.json
2019-08-20 13:57:14,462 - brc - INFO - Loss on dev set: 15.376707763671876
2019-08-20 13:57:14,462 - brc - INFO - Result on dev set: {'Bleu-4': 0.012630279580145154, 'Rouge-L': 0.1032146129414365}
2019-08-20 13:57:14,462 - brc - INFO - Predicted answers are saved to ../data/results/
2019-08-20 13:57:14,462 - brc - INFO - Predicting answers for test set...
(0, 19) [13934, 17433, 14253, 14353, 8721, 17340, 2882, 1, 8091, 14890, 6634, 9692, 8387, 12072, 12879, 17433, 2958, 13757, 1430, 11440] 孙中山是有个儿子的叫<unk>,当过行政院长宋庆龄至少是第二任了
(0, 0) [1484] 将
(0, 0) [2271] 交换机
(0, 0) [16982] 不
(0, 0) [6659] 来自
(0, 0) [10219] 确定
(0, 38) [11310, 12598, 13154, 8161, 10148, 8335, 6178, 9771, 8198, 17340, 742, 1, 3755, 3234, 6091, 16163, 5113, 12598, 7630, 12632, 3159, 8091, 12327, 12362, 2427, 742, 1, 17255, 16163, 5113, 12598, 7630, 12632, 3159, 8091, 12327, 17670, 2427, 742] 违法行为:违反禁令标志、禁止标线指示的。<unk>2013年新处罚标准:罚款100元,扣3分。<unk>原处罚标准:罚款100元,扣2分。
(0, 0) [1765] 凯乐石
(0, 5) [13866, 13512, 14432, 2558, 14302, 1670] 痛风能吃什么菜啊
(0, 0) [4874] 高山
(0, 0) [3210] 您好
(0, 8) [8933, 16460, 13693, 16564, 11071, 11896, 17340, 5010, 17433] 下列不能缓解和消除肌肉酸痛的方法是
(0, 10) [4140, 10414, 15578, 6148, 6634, 2148, 17106, 4517, 8091, 17770, 15770] 账号被举报次数过多存在异常,就是这样!
(36, 36) [4215] 喝完
(0, 6) [3516, 5779, 2451, 2756, 14253, 3516, 1] 可以换点徽章有可以<unk>
(0, 0) [12405] 贪玩
(0, 0) [14072] 哪个
(0, 28) [2795, 11411, 8091, 4176, 7561, 15841, 4778, 7896, 13899, 2222, 17340, 8091, 15738, 6859, 14302, 16790, 5880, 1022, 16363, 14253, 4737, 17340, 8091, 14760, 3439, 17340, 16611, 4664, 742] 美食天下,我也是不会做饭就看这个网站的,每道菜都是上面图下面有步骤的,很详细的介绍第一步。
(70, 70) [742] 。
(0, 64) [9923, 17433, 8222, 11319, 17340, 619, 1, 8162, 8091, 14253, 1912, 8162, 8335, 17714, 1, 8162, 8335, 3985, 1, 8335, 8222, 14251, 9790, 2567, 12795, 1029, 8091, 16609, 14253, 13220, 14253, 1, 1074, 14253, 1328, 17340, 1, 14251, 9790, 2567, 14268, 742, 16363, 16796, 14022, 4272, 9044, 4059, 16737, 11440, 1328, 7965, 8579, 3937, 17340, 11836, 8091, 798, 9044, 4059, 17340, 1328, 17409, 1, 742] 阳江市是中国著名的旅游<unk>城市,有平安城市、最佳<unk>城市、国家<unk>、中国风筝之乡等荣誉,素有北有<unk>南有阳江的<unk>风筝之乡称谓。下面百分网小编为大家整理了阳江一日游最受欢迎的旅游景点,希望能够为大家的阳江旅行<unk>。
(0, 0) [14136] 使用
(0, 0) [7369] 少儿
(0, 4) [6489, 17340, 11522, 7896, 5399] 长的好看就行
(0, 0) [5126] 不知道
(49, 49) [742] 。
(0, 0) [14586] 任何
(0, 0) [15847] 南怀谨
(0, 0) [11905] 共
(0, 6) [17298, 16790, 946, 14839, 1, 17340, 742] 爷爷都是从孙子<unk>的。
(0, 0) [947] 1
(0, 3) [3158, 7961, 5694, 5694] 可是纱窗……
(0, 10) [4920, 8805, 6723, 5669, 2328, 8091, 15898, 17563, 8107, 17433, 17593] ?心脏支架手术后,一般存活寿命是多少
(0, 28) [14298, 5741, 6652, 288, 1, 8091, 14379, 10912, 17125, 14322, 12598, 11142, 1493, 12700, 11713, 12598, 3416, 11142, 8335, 12502, 8335, 1493, 8091, 6652, 17231, 9044, 5099, 11958, 742] 营业税改增值税代码<unk>,应税项目范围:经纪代理服务;包括:各类经纪、中介、代理服务,增值税税率为6%。
(0, 5) [9621, 12613, 13712, 947, 15243, 11854] 我的世界1.8.1免费mac
(0, 0) [440] 别
(0, 0) [10681] 小规模
(0, 0) [12163] 12306
(0, 0) [947] 1
(0, 0) [1] <unk>
(0, 0) [14369] 应该
(0, 0) [14761] 按
(0, 24) [14205, 12011, 12598, 4434, 6995, 7023, 8335, 5688, 2610, 12802, 17340, 16433, 3842, 13759, 5037, 13183, 17340, 9934, 6258, 5299, 7585, 17340, 16295, 13024, 742] 准备证件:双方身份证原件、集体户口簿内的户口卡或者加盖单位印章的记载其户籍情况的户口簿复印件。
(0, 14) [12086, 14260, 7998, 8185, 11401, 742, 14369, 17433, 12598, 261, 16363, 9515, 14353, 17433, 1194] 题目好像没写正确。应该是:【下面两个是公式
(0, 0) [7774] 这
(0, 10) [920, 3607, 9037, 17340, 1, 7626, 16564, 920, 3607, 9037, 13063] 减数第一次分裂的<unk>时期和减数第一次分裂后期
(0, 13) [7203, 11316, 11316, 7896, 7803, 11440, 8091, 7940, 17340, 12466, 485, 6495, 13486, 11166] 每个试试就知道了,不行的话要调整表格大小
(0, 19) [6721, 17433, 6193, 15977, 1018, 11958, 17337, 17340, 8775, 8091, 11237, 17433, 6193, 15977, 11089, 13712, 11458, 17337, 17340, 8775] s990是指含有99%以上的银,s925是指含有92.5以上的银
(0, 8) [5351, 2756, 5104, 4977, 10022, 12598, 1133, 1, 7353] 时光徽章兑换余额帖:[<unk>]
(0, 0) [4264] 健康
(0, 0) [12664] 因为
(0, 0) [14691] 你
(0, 0) [10738] 邮政
(0, 5) [17743, 4618, 17433, 6474, 16559, 4920] 古力娜扎是穆斯林吗?
(0, 21) [1, 7410, 14253, 13015, 1, 8091, 1, 9790, 5401, 742, 13148, 16835, 7410, 1, 3040, 8091, 16401, 12895, 9790, 5401, 742, 13148] <unk>(有翱翔<unk>,<unk>之意。)凯瑞(<unk>高贵,潇洒大气之意。)
(0, 0) [4920] ?
(0, 4) [1460, 10695, 12762, 14784, 975] 上海石油化工交易中心***
(0, 0) [11553] 这个
(0, 0) [14496] 没有
(0, 0) [17376] 王者
(0, 4) [2558, 17433, 9515, 15378, 3586] 什么是两参人员
(0, 0) [14392] 还有
(0, 174) [7998, 8403, 530, 7139, 14691, 17340, 1, 14022, 8091, 11273, 15555, 11601, 12183, 8497, 8091, 15531, 15550, 6337, 17496, 9722, 8359, 2908, 7244, 8091, 9490, 17433, 2148, 14022, 2211, 3281, 15682, 8091, 14691, 12346, 17340, 3172, 14840, 7090, 4868, 8497, 4278, 7998, 12532, 7561, 3172, 14840, 8091, 1367, 7774, 4887, 4887, 17340, 7886, 15628, 1, 6224, 15550, 6337, 17489, 14691, 7815, 1, 17340, 7815, 14840, 17340, 12446, 8091, 13609, 3281, 15786, 8497, 4190, 16322, 5854, 3130, 7839, 12325, 17340, 8091, 7184, 6933, 8497, 10345, 12977, 13566, 17433, 485, 3810, 9802, 1, 17433, 9802, 9195, 2225, 8091, 12041, 12041, 3386, 1, 17340, 12446, 742, 4868, 8497, 15898, 6337, 14253, 9619, 17340, 1, 8091, 5238, 9809, 9619, 6997, 3281, 15841, 5216, 742, 8950, 8497, 14496, 10414, 13357, 13012, 10871, 17433, 14691, 5265, 16564, 11273, 7016, 1, 8091, 11273, 17340, 1, 5186, 13012, 8764, 12852, 8091, 1, 6933, 10414, 1, 17340, 3281, 3516, 11316, 4489, 12316, 12316, 11204, 17340, 6224, 742, 14162, 6211, 8091, 7628, 2564, 17340, 13341, 13610, 13497, 9802, 7401, 6982, 3130, 8091, 15980, 1, 742] 没听说过只能说明你的<unk>小,萌芽每天收到无数投稿,却只会在每个月登几篇,比例是多小大概也清楚,你所谓的多数人反映邮箱投稿根本没回音也是多数人,毕竟这薄薄的一本<unk>杂志只会选取你少数<unk>的少数人的文章,原先也有人投稿成功然后来吧里炫耀的,私人觉得投稿失败这种事情首要是要让自己<unk>是自己不够好,对比对比已<unk>的文章。邮箱投稿一般会有系统的<unk>,不过很多时候系统繁忙也不会回复。再者投稿没有被录取也有可能是你文风和萌芽不太<unk>,萌芽的<unk>里面也有提到这一点,<unk>觉得被<unk>的也可以试着投投其他的杂志。如果还是不行,就像我说的还是先提高自己写作水平吧,望<unk>。
(0, 10) [15573, 4584, 17340, 4090, 2808, 10871, 14969, 4920, 12548, 1061, 4920] 作为闲鱼的卖家有没有可能被骗?怎么骗?
(0, 0) [15674] 徐州
(0, 0) [1] <unk>
(0, 1) [15297, 17340] 夏威夷果的
(0, 0) [13989] 成品
(0, 1) [14764, 17340] 现代的
(0, 0) [1] <unk>
(0, 0) [272] 工伤保险
(0, 5) [2259, 17340, 11325, 7292, 2225, 3304] trilogy的玫瑰精油特别好用
(0, 0) [11210] 过期
(0, 0) [17433] 是
(0, 0) [3860] 更多
(0, 11) [16154, 9837, 2379, 16524, 12304, 14299, 1415, 2495, 5204, 5010, 16886, 12598] 卡西欧手表ga-11024小时制调试方法如图:
(0, 0) [10972] 安装
(0, 0) [4264] 健康
(0, 0) [947] 1
(0, 0) [7206] serial
(0, 20) [9100, 17518, 13610, 2427, 14070, 7745, 17340, 947, 16554, 17670, 8091, 12459, 17340, 947, 16554, 17670, 17440, 17762, 5672, 1, 742] 另一方先分得共同财产的1/2,余下的1/2由所有继承人<unk>。
(0, 0) [947] 1
(0, 0) [2958] 第
(0, 0) [261] 【
(0, 3) [2202, 8091, 1368, 8091] 龙珠,龙珠z,
(0, 0) [9918] 近日
(0, 0) [7203] 每个
(0, 0) [3263] gta5
(0, 0) [14691] 你
(0, 0) [947] 1
(0, 0) [16242] 碎片
(0, 4) [14761, 4515, 7994, 14761, 10156] 按0再按拨号
(0, 77) [4579, 6658, 2801, 17340, 17381, 530, 17496, 409, 12802, 9181, 8091, 14676, 11440, 409, 17433, 7663, 9181, 17340, 8091, 9181, 4737, 12598, 947, 13712, 9191, 2973, 4257, 4579, 742, 17670, 13712, 2021, 6658, 4783, 5224, 17340, 16264, 9308, 742, 12362, 13712, 10917, 14157, 17381, 8091, 7981, 9181, 8009, 7482, 742, 13915, 13712, 12764, 9181, 3579, 742, 3664, 12598, 2204, 485, 9181, 17340, 7033, 8091, 14162, 16631, 14676, 17670, 13739, 17340, 17381, 8091, 17433, 16460, 10414, 9181, 17340, 742] 微信发出去的消息只能在两分钟内撤回,超过了两分钟是无法撤回的,撤回步骤:1.打开并登陆微信。2.进入发照片好友的聊天界面。3.长按该消息,会出现撤回提示选项。4.点击撤回即可。注意:对于要撤回的信息,如果发送超过2分钟的消息,是不能被撤回的。
(0, 6) [9958, 645, 10748, 1121, 2227, 10658, 12598] 西安三星冰箱售后维修中心许诺:
(0, 1) [4098, 16400] 去正规
(0, 0) [3804] 世界上
(0, 0) [1] <unk>
(0, 8) [642, 10179, 1, 1133, 947, 7353, 742, 6390, 16900] 如何制定<unk>[1]。部门目标
(0, 0) [14375] 成语大全
(0, 0) [1] <unk>
(0, 0) [947] 1
2019-08-20 13:57:17,044 - brc - INFO - Saving test.predicted results to ../data/results/test.predicted.json

最起码可以进行完整训练,以及部分答案可读性是可以的

@fooSynaptic
Copy link

不过我有个问题,rep主提到的self.attention部分的实现为何我没有找到,难道rep主把attention flow当做了self-attention?

@shiningliang
Copy link
Owner

不过我有个问题,rep主提到的self.attention部分的实现为何我没有找到,难道rep主把attention flow当做了self-attention?

是的是的,在attention flow中加了一种self attention,跟现行的基于transformer的self attention不一样

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants