saattrupdan/icebert-texas-squad-is

TExAS-SQuAD-is

This model is a fine-tuned version of IceBERT on the TExAS-SQuAD-is dataset. It achieves the following results on the evaluation set:

• Exact match: xx.xx%
• F1-score: xx.xx%

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
2.5353	0.12	500	2.2356
2.364	0.24	1000	2.0607
2.2243	0.36	1500	2.0617
2.1403	0.49	2000	1.9934
2.1491	0.61	2500	2.0515
2.0604	0.73	3000	1.9602
2.0232	0.85	3500	1.8954
2.0905	0.97	4000	1.9474
1.9229	1.09	4500	1.9814
1.9162	1.22	5000	1.9053
1.8937	1.34	5500	1.9501
1.9085	1.46	6000	1.8882
1.8671	1.58	6500	1.8996
1.8997	1.7	7000	1.8340
1.8546	1.82	7500	1.8883
1.8935	1.95	8000	1.8567
1.7031	2.07	8500	1.9206
1.7699	2.19	9000	1.8790
1.7016	2.31	9500	1.8670
1.7744	2.43	10000	1.8951
1.7518	2.55	10500	1.9550
1.7503	2.68	11000	1.9120
1.7818	2.8	11500	1.8820
1.6955	2.92	12000	1.8908

Framework versions

• Transformers 4.12.2
• Pytorch 1.8.1+cu101
• Datasets 1.12.1
• Tokenizers 0.10.3