bart-base-asqa-cb

This model is a fine-tuned version of facebook/bart-base on the ASQA dataset. It achieves the following results on the evaluation set:

• Loss: 2.7878
• Rougelsum: 36.5701

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rougelsum
No log	1.0	273	2.9082	35.2452
3.4369	2.0	546	2.8642	35.9217
3.4369	3.0	819	2.8426	35.9304
3.1616	4.0	1092	2.8310	36.2562
3.1616	5.0	1365	2.8193	36.4633
3.0814	6.0	1638	2.8091	36.6044
3.0814	7.0	1911	2.8069	36.6191
3.0165	8.0	2184	2.8026	36.6380
3.0165	9.0	2457	2.7978	36.6962
2.9724	10.0	2730	2.7965	36.5703
2.9282	11.0	3003	2.7926	36.5339
2.9282	12.0	3276	2.7916	36.5093
2.8996	13.0	3549	2.7911	36.4693
2.8996	14.0	3822	2.7904	36.3852
2.8803	15.0	4095	2.7888	36.6173
2.8803	16.0	4368	2.7881	36.5282
2.8653	17.0	4641	2.7885	36.6131
2.8653	18.0	4914	2.7878	36.6120
2.8558	19.0	5187	2.7877	36.5637
2.8558	20.0	5460	2.7878	36.5701

Framework versions

• Transformers 4.23.0.dev0
• Pytorch 1.12.1+cu102
• Datasets 2.4.0
• Tokenizers 0.12.1