train-bart-base

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

• Loss: 0.2710
• Rouge1: 42.8665
• Rouge2: 21.8559
• Rougel: 37.536
• Rougelsum: 39.3725
• Gen Len: 18.0

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-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
0.3316	1.0	1557	0.2421	41.223	19.5022	35.5882	38.1294	18.0
0.2448	2.0	3115	0.2304	41.9635	20.5356	36.729	38.7748	18.0
0.2088	3.0	4672	0.2317	41.1639	20.168	35.9644	38.0607	18.0
0.1811	4.0	6230	0.2352	42.5001	21.4806	37.0514	39.0242	18.0
0.1591	5.0	7787	0.2422	42.148	20.9001	36.7976	38.6102	18.0
0.1399	6.0	9345	0.2465	42.1862	21.1403	36.7742	38.7401	18.0
0.1247	7.0	10902	0.2535	42.8571	21.998	37.6668	39.5963	18.0
0.1115	8.0	12460	0.2609	42.2841	21.1273	36.9562	38.9423	18.0
0.1019	9.0	14017	0.2677	42.8866	21.6628	37.5422	39.4627	18.0
0.0946	10.0	15570	0.2710	42.8665	21.8559	37.536	39.3725	18.0

Framework versions

• Transformers 4.39.3
• Pytorch 2.1.2
• Datasets 2.19.1
• Tokenizers 0.15.2