加载预训练模型出错

[DoubleSsh]

运行命令为python train.py --embedding --model_dir model_save_dir --dataset okvqa --validate 报错：Some weights of the model checkpoint at unc-nlp/lxmert-base-uncased were not used when initializing LxmertModel: ['obj_predict_head.decoder_dict.attr.weight', 'cls.predictions.decoder.weight', 'cls.predictions.bias', 'obj_predict_head.decoder_dict.obj.weight', 'obj_predict_head.decoder_dict.attr.bias', 'answer_head.logit_fc.2.weight', 'answer_head.logit_fc.3.weight', 'obj_predict_head.decoder_dict.obj.bias', 'answer_head.logit_fc.0.weight', 'obj_predict_head.transform.dense.weight', 'cls.predictions.transform.dense.weight', 'obj_predict_head.transform.LayerNorm.bias', 'cls.predictions.transform.LayerNorm.weight', 'answer_head.logit_fc.3.bias', 'obj_predict_head.transform.dense.bias', 'cls.seq_relationship.weight', 'cls.seq_relationship.bias', 'answer_head.logit_fc.0.bias', 'obj_predict_head.decoder_dict.feat.bias', 'answer_head.logit_fc.2.bias', 'cls.predictions.transform.dense.bias', 'obj_predict_head.decoder_dict.feat.weight', 'obj_predict_head.transform.LayerNorm.weight', 'cls.predictions.transform.LayerNorm.bias']

• This IS expected if you are initializing LxmertModel from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
• This IS NOT expected if you are initializing LxmertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model). 9009 已杀死 请问这是什么原因呢？预训练模型（unc-nlp/lxmert-base-uncased）我是有下载的。在linux上运行的。

[tanshaosuannai]

大佬，我想请问一下这个unc-nlp/lxmert-base-uncased预训练模型在哪下载啊。

[tanshaosuannai]

运行命令为python train.py --embedding --model_dir model_save_dir --dataset okvqa --validate 报错：初始化 LxmertModel 时未使用 unc-nlp/lxmert-base-uncased 模型检查点的某些权重： ['obj_predict_head.decoder_dict.attr.weight'， 'cls.predictions.decoder.weight'， 'cls.predictions.bias'， 'obj_predict_head.decoder_dict.obj.weight'， 'obj_predict_head.decoder_dict.attr.bias'， 'answer_head.logit_fc.2.weight'， 'answer_head.logit_fc.3.weight'， 'obj_predict_head.decoder_dict.obj.bias'， 'answer_head.logit_fc.0.weight'， 'obj_predict_head.transform.dense.weight'， 'cls.predictions.transform.dense.weight'， 'obj_predict_head.transform.LayerNorm.bias'， 'cls.predictions.transform.LayerNorm.weight'， 'answer_head.logit_fc.3.bias'， 'obj_predict_head.transform.dense.bias'， 'cls.seq_relationship.weight'， 'cls.seq_relationship.bias'， 'answer_head.logit_fc.0.bias'， 'obj_predict_head.decoder_dict.feat.bias'， 'answer_head.logit_fc.2.bias'， 'cls.predictions.transform.dense.bias'， 'obj_predict_head.decoder_dict.feat.weight'， 'obj_predict_head.transform.LayerNorm.weight'， 'cls.predictions.transform.LayerNorm.bias']

• 如果要从在另一个任务上训练的模型的检查点初始化 LxmertModel，或者使用其他体系结构（例如，从 BertForPreTraining 模型初始化 BertForSequenceClassification 模型），则这是预期的。
• 如果从预期完全相同的模型的检查点初始化 LxmertModel（从 BertForSequenceClassification 模型初始化 BertForSequenceClassification 模型），则不需要这样做。 9009 已杀死 请问这是什么原因呢？预训练模型（unc-nlp/lxmert-base-uncased）我是有下载的。在linux上运行的。

大佬，我想请问一下这个unc-nlp/lxmert-base-uncased预训练模型在哪下载啊。