finetuned-BART-UK-financial-summarization

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

• Loss: 1.8235
• Rouge1: 43.0491
• Rouge2: 31.1442
• Rougel: 37.3232
• Rougelsum: 39.8716
• Gen Len: 121.2402

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: 1
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 4.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
2.2823	0.34	1000	2.1138	13.157	8.2441	11.8879	12.1511	20.0
1.8952	0.68	2000	1.9969	13.4743	8.9624	12.214	12.605	20.0
1.7594	1.02	3000	1.9643	14.0421	9.6907	12.9115	13.141	20.0
1.5823	1.36	4000	1.9357	14.717	10.5958	13.6169	13.9246	20.0
1.4926	1.7	5000	1.8886	14.7609	10.8361	13.7826	14.0178	19.9777
1.5201	2.04	6000	1.9262	14.9849	11.0313	14.0855	14.3163	20.0
1.3739	2.38	7000	1.8690	15.3066	11.2232	14.2604	14.6091	20.0
1.3481	2.72	8000	1.8463	15.0157	11.0947	14.0939	14.3456	20.0
1.2094	3.06	9000	1.8442	15.0272	10.8948	13.9635	14.3162	20.0
1.0402	3.4	10000	1.8370	15.1651	11.2822	14.2048	14.4765	19.9888
1.1247	3.74	11000	1.8335	15.271	11.3492	14.3258	14.5952	20.0

Framework versions

• Transformers 4.16.0.dev0
• Pytorch 1.7.0
• Datasets 1.17.0
• Tokenizers 0.11.0