# 'BertEncoder' object has no attribute 'gradient\_checkpointing' **URL:** \ **Category:** 🤗Transformers\ **Created:** [October 29, 2021, 10:34pm UTC](https://discuss.huggingface.co/t/bertencoder-object-has-no-attribute-gradient-checkpointing/11207 "2021-10-29T22:34:40Z")\ **Posts on this page:** 3\ **Page:** 1
**Author:** ![GullyBurns](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/gullyburns/32/29626_2.png) [@GullyBurns](https://discuss.huggingface.co/u/GullyBurns)\ **Post date:** [October 29, 2021, 10:34pm UTC](https://discuss.huggingface.co/t/bertencoder-object-has-no-attribute-gradient-checkpointing/11207/1 "2021-10-29T22:34:40Z")
I’m getting a strange error that previously worked OK. I’m only trying to use a previously trained NLP model to predict a label. Source Code: ```auto BATCH_SIZE = 128 MAX_LEN = 64 def generate_labels(model, tokenizer, df, segment_label): segments = list(df[segment_label].values) pred_labels = [] num_batches = int(len(segments) / BATCH_SIZE) + 1 print('Total num batches: ', num_batches) model.eval() with torch.no_grad(): for i in range(num_batches): if (i + 1) % 100 == 0: print('Processing batch #', (i + 1)) batch = segments[i * BATCH_SIZE : (i + 1) * BATCH_SIZE] tokenized_text = tokenizer(batch, return_tensors="pt", padding = 'max_length', \ truncation = True, max_length = MAX_LEN) input_ids = tokenized_text['input_ids'].to('cuda') attention_mask = tokenized_text['attention_mask'].to('cuda') outputs = model(input_ids, attention_mask = attention_mask).logits pred_labels_batch = torch.argmax(outputs, dim = 1).cpu().numpy() pred_labels.extend(pred_labels_batch) return pred_labels`Preformatted text` pred_labels = generate_labels(trained_model, tokenizer, sent_df, 'text') ``` ERROR: ```auto --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in ----> 1 pred_labels = generate_labels(trained_model, tokenizer, sent_df, 'text') 2 sent_df['PREDICTED_DISCOURSE_TAG'] = [pubmed_rct_idx_to_label[x] for x in pred_labels] in generate_labels(model, tokenizer, df, segment_label) 17 input_ids = tokenized_text['input_ids'].to('cuda') 18 attention_mask = tokenized_text['attention_mask'].to('cuda') ---> 19 outputs = model(input_ids, attention_mask = attention_mask).logits 20 pred_labels_batch = torch.argmax(outputs, dim = 1).cpu().numpy() 21 pred_labels.extend(pred_labels_batch) /databricks/python/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs) 1049 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks 1050 or _global_forward_hooks or _global_forward_pre_hooks): -> 1051 return forward_call(*input, **kwargs) 1052 # Do not call functions when jit is used 1053 full_backward_hooks, non_full_backward_hooks = [], [] /databricks/python/lib/python3.8/site-packages/transformers/models/bert/modeling_bert.py in forward(self, input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, labels, output_attentions, output_hidden_states, return_dict) 1528 return_dict = return_dict if return_dict is not None else self.config.use_return_dict 1529 -> 1530 outputs = self.bert( 1531 input_ids, 1532 attention_mask=attention_mask, /databricks/python/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs) 1049 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks 1050 or _global_forward_hooks or _global_forward_pre_hooks): -> 1051 return forward_call(*input, **kwargs) 1052 # Do not call functions when jit is used 1053 full_backward_hooks, non_full_backward_hooks = [], [] /databricks/python/lib/python3.8/site-packages/transformers/models/bert/modeling_bert.py in forward(self, input_ids, attention_mask, token_type_ids, position_ids, head_mask, inputs_embeds, encoder_hidden_states, encoder_attention_mask, past_key_values, use_cache, output_attentions, output_hidden_states, return_dict) 994 past_key_values_length=past_key_values_length, 995 ) --> 996 encoder_outputs = self.encoder( 997 embedding_output, 998 attention_mask=extended_attention_mask, /databricks/python/lib/python3.8/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs) 1049 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks 1050 or _global_forward_hooks or _global_forward_pre_hooks): -> 1051 return forward_call(*input, **kwargs) 1052 # Do not call functions when jit is used 1053 full_backward_hooks, non_full_backward_hooks = [], [] /databricks/python/lib/python3.8/site-packages/transformers/models/bert/modeling_bert.py in forward(self, hidden_states, attention_mask, head_mask, encoder_hidden_states, encoder_attention_mask, past_key_values, use_cache, output_attentions, output_hidden_states, return_dict) 558 past_key_value = past_key_values[i] if past_key_values is not None else None 559 --> 560 if self.gradient_checkpointing and self.training: 561 562 if use_cache: /databricks/python/lib/python3.8/site-packages/torch/nn/modules/module.py in __getattr__ (self, name) 1128 if name in modules: 1129 return modules[name] -> 1130 raise AttributeError("'{}' object has no attribute '{}'".format( 1131 type(self). __name__ , name)) 1132 AttributeError: 'BertEncoder' object has no attribute 'gradient_checkpointing' ``` No idea where this is coming from. Any ideas or help? Note that this error goes away completely if I use `transformers==4.10.0` ---
**Author:** ![sgugger](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/sgugger/32/2291_2.png) [@sgugger](https://discuss.huggingface.co/u/sgugger)\ **Post date:** [October 31, 2021, 3:33pm UTC](https://discuss.huggingface.co/t/bertencoder-object-has-no-attribute-gradient-checkpointing/11207/2 "2021-10-31T15:33:00Z")
You did not show how you create your model in your post, so no one can help you debug the problem. Form the error message, it looks like you used `torch.save` to save your whole model (and not the weights), which is not recommended at all because when the model changes (like it did between 4.10 and 4.11) you then can’t reload it directly with `torch.load`. Our advice is to always use `save_pretrained`/`from_pretrained` to save/load your models or if it’s not possible, to save the weights (`model.state_dict`) with `torch.save` and then reload them with `model.load_state_dict`, as this will works across different versions of the models. ---
**Author:** ![Thang](https://avatars.discourse-cdn.com/v4/letter/t/8edcca/32.png) [@Thang](https://discuss.huggingface.co/u/Thang)\ **Post date:** [August 1, 2022, 5:58am UTC](https://discuss.huggingface.co/t/bertencoder-object-has-no-attribute-gradient-checkpointing/11207/3 "2022-08-01T05:58:10Z")
The problem is from Torch and Transformer. Try to test with different versions of them. In my case, I installed Transformers==4.0.0 and the problem was solved.