--- base_model: google-bert/bert-base-uncased library_name: peft license: apache-2.0 metrics: - accuracy tags: - trl - sft - generated_from_trainer model-index: - name: bert-base-uncased-swag results: [] --- # bert-base-uncased-swag This model is a fine-tuned version of [google-bert/bert-base-uncased](https://huggingface.co/google-bert/bert-base-uncased) on [SWAG](https://huggingface.co/datasets/allenai/swag) dataset. It achieves the following results on the evaluation set: - Loss: 0.6749 - Accuracy: 0.7503 ## Model description ## Intended uses & limitations This model should be used as an expert in the [Meteor-of-LoRA framework](https://github.com/ParagonLight/meteor-of-lora). ## Training and evaluation data The data were splitted based on HuggingFace default dataset: ```python3 dataset = load_dataset("swag") ``` ## Training procedure Our approach focuses explicitly on adapting the Transformers weights' Wq (query) and Wv (value) in the attention module for parameter efficiency. ### Training hyperparameters The following hyperparameters were used during training: - learning_rate: 5e-05 - train_batch_size: 16 - eval_batch_size: 16 - seed: 42 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: linear - num_epochs: 5 ### Training results | Training Loss | Epoch | Step | Validation Loss | Accuracy | |:-------------:|:------:|:-----:|:---------------:|:--------:| | 1.3807 | 0.1088 | 500 | 1.2507 | 0.6138 | | 1.1949 | 0.2175 | 1000 | 1.0938 | 0.5737 | | 1.132 | 0.3263 | 1500 | 1.0330 | 0.5657 | | 1.0348 | 0.4351 | 2000 | 0.9162 | 0.6440 | | 1.0008 | 0.5438 | 2500 | 0.8464 | 0.6801 | | 0.9609 | 0.6526 | 3000 | 0.8267 | 0.6859 | | 0.9454 | 0.7614 | 3500 | 0.8116 | 0.6943 | | 0.9512 | 0.8701 | 4000 | 0.8125 | 0.6955 | | 0.9367 | 0.9789 | 4500 | 0.7838 | 0.7032 | | 0.9205 | 1.0877 | 5000 | 0.7861 | 0.7044 | | 0.9189 | 1.1964 | 5500 | 0.7713 | 0.7088 | | 0.8975 | 1.3052 | 6000 | 0.7538 | 0.7173 | | 0.9065 | 1.4140 | 6500 | 0.7520 | 0.7175 | | 0.8957 | 1.5227 | 7000 | 0.7513 | 0.7200 | | 0.8768 | 1.6315 | 7500 | 0.7411 | 0.7195 | | 0.8858 | 1.7403 | 8000 | 0.7306 | 0.7262 | | 0.875 | 1.8490 | 8500 | 0.7302 | 0.7268 | | 0.8649 | 1.9578 | 9000 | 0.7229 | 0.7303 | | 0.8653 | 2.0666 | 9500 | 0.7126 | 0.7322 | | 0.867 | 2.1753 | 10000 | 0.7198 | 0.7293 | | 0.868 | 2.2841 | 10500 | 0.7125 | 0.7346 | | 0.855 | 2.3929 | 11000 | 0.7051 | 0.7350 | | 0.8557 | 2.5016 | 11500 | 0.7008 | 0.7384 | | 0.8622 | 2.6104 | 12000 | 0.6979 | 0.7389 | | 0.8506 | 2.7192 | 12500 | 0.7068 | 0.7378 | | 0.8558 | 2.8279 | 13000 | 0.7082 | 0.7337 | | 0.849 | 2.9367 | 13500 | 0.6978 | 0.7407 | | 0.8581 | 3.0455 | 14000 | 0.6850 | 0.7460 | | 0.8521 | 3.1542 | 14500 | 0.6945 | 0.7428 | | 0.8454 | 3.2630 | 15000 | 0.6863 | 0.7446 | | 0.8257 | 3.3718 | 15500 | 0.6917 | 0.7414 | | 0.8522 | 3.4805 | 16000 | 0.6882 | 0.7445 | | 0.8359 | 3.5893 | 16500 | 0.6845 | 0.7442 | | 0.8238 | 3.6981 | 17000 | 0.6863 | 0.7441 | | 0.8382 | 3.8068 | 17500 | 0.6937 | 0.7438 | | 0.8326 | 3.9156 | 18000 | 0.6780 | 0.7488 | | 0.8344 | 4.0244 | 18500 | 0.6775 | 0.7484 | | 0.8224 | 4.1331 | 19000 | 0.6811 | 0.7477 | | 0.8261 | 4.2419 | 19500 | 0.6797 | 0.7480 | | 0.8256 | 4.3507 | 20000 | 0.6815 | 0.7481 | | 0.8191 | 4.4594 | 20500 | 0.6788 | 0.7476 | | 0.838 | 4.5682 | 21000 | 0.6802 | 0.7490 | | 0.8383 | 4.6770 | 21500 | 0.6753 | 0.7498 | | 0.8343 | 4.7857 | 22000 | 0.6762 | 0.7498 | | 0.8381 | 4.8945 | 22500 | 0.6749 | 0.7503 | ### Framework versions - PEFT 0.12.1.dev0 - Transformers 4.45.0.dev0 - Pytorch 2.3.1+cu121 - Datasets 2.21.0 - Tokenizers 0.19.1