| 18:29:41 {'seed': 1, 'ver': 'v1d', 'use_log': True, 'use_tqdm': True, 'debug': False, 'tokenizer': BertTokenizerFast(name_or_path='yiyanghkust/finbert-tone', vocab_size=30873, model_max_length=1000000000000000019884624838656, is_fast=True, padding_side='right', truncation_side='right', special_tokens={'unk_token': '[UNK]', 'sep_token': '[SEP]', 'pad_token': '[PAD]', 'cls_token': '[CLS]', 'mask_token': '[MASK]'}, clean_up_tokenization_spaces=True), 'config': <custom_config.LongBERTConfig object at 0x7f173b965890>, 'max_len': 50000, 'train_one_part': False, 'gradient_accumulation_steps': 2, 'apex': True, 'device': device(type='cuda', index=1), 'nepochs': 10, 'batch_size': 2, 'num_workers': 128, 'lr': 2e-05, 'weight_decay': 0.01, 'encoder_lr': 2e-05, 'decoder_lr': 0.001, 'min_lr': 1e-06, 'eps': 1e-06, 'betas': (0.9, 0.999), 'scheduler_type': 'cosine', 'num_cycles': 0.5, 'num_warmup_steps': 0.0, 'train_data_dir': 'data/train', 'valid_data_dir': 'data/valid', 'test_data_dir': '.', 'output_dir': 'model/v1/d'} | |
| 18:29:41 Preparing training materials... | |
| 18:29:41 Preparing the model... | |
| 18:29:41 loading weights file pytorch_model.bin from cache at /home/tphan/.cache/huggingface/hub/models--yiyanghkust--finbert-tone/snapshots/4921590d3c0c3832c0efea24c8381ce0bda7844b/pytorch_model.bin | |
| 18:29:42 Some weights of the model checkpoint at yiyanghkust/finbert-tone were not used when initializing BertModel: ['classifier.weight', 'classifier.bias'] | |
| - This IS expected if you are initializing BertModel 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 BertModel from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model). | |
| 18:29:42 All the weights of BertModel were initialized from the model checkpoint at yiyanghkust/finbert-tone. | |
| If your task is similar to the task the model of the checkpoint was trained on, you can already use BertModel for predictions without further training. | |
| 18:29:44 Preparing the dataloaders... | |
| 13:39:11 Epoch: [1] - Train/Valid Loss: 5.0073/4.6463 | |
| 13:39:11 Saving the model to model/v1/d | |
| 08:47:46 Epoch: [2] - Train/Valid Loss: 4.4927/4.2638 | |
| 08:47:46 Saving the model to model/v1/d | |
| 03:57:45 Epoch: [3] - Train/Valid Loss: 4.1972/4.0526 | |
| 03:57:45 Saving the model to model/v1/d | |
| 23:09:42 Epoch: [4] - Train/Valid Loss: 3.9879/3.8690 | |
| 23:09:42 Saving the model to model/v1/d | |
| 18:15:12 Epoch: [5] - Train/Valid Loss: 3.8552/3.7749 | |
| 18:15:12 Saving the model to model/v1/d | |
| 13:20:43 Epoch: [6] - Train/Valid Loss: 3.7773/3.7123 | |
| 13:20:43 Saving the model to model/v1/d | |
| 08:26:20 Epoch: [7] - Train/Valid Loss: 3.7244/3.6654 | |
| 08:26:20 Saving the model to model/v1/d | |
| 03:32:07 Epoch: [8] - Train/Valid Loss: 3.6823/3.6315 | |
| 03:32:07 Saving the model to model/v1/d | |