patent-summarization-fb-bart-base-2022-09-20

This model is a fine-tuned version of facebook/bart-base on the farleyknight/big_patent_5_percent dataset. It achieves the following results on the evaluation set:

• Loss: 2.4088
• Rouge1: 39.4401
• Rouge2: 14.2445
• Rougel: 26.2701
• Rougelsum: 33.7535
• Gen Len: 78.9702

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: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
3.0567	0.08	5000	2.8864	18.9387	7.1014	15.4506	16.8377	19.9979
2.9285	0.17	10000	2.7800	19.8983	7.3258	16.0823	17.7019	20.0
2.9252	0.25	15000	2.7080	19.6623	7.4627	16.0153	17.4485	20.0
2.8123	0.33	20000	2.6585	19.7414	7.5251	15.8166	17.4668	20.0
2.7117	0.41	25000	2.6070	19.7661	7.7193	16.2795	17.7884	20.0
2.7131	0.5	30000	2.5616	19.6706	7.4229	15.7998	17.4324	20.0
2.6373	0.58	35000	2.5250	20.0155	7.6811	16.1231	17.7578	20.0
2.6785	0.66	40000	2.4977	20.0974	7.9578	16.543	18.0242	20.0
2.6265	0.75	45000	2.4701	19.994	7.9114	16.3501	17.8786	20.0
2.5833	0.83	50000	2.4441	19.9981	7.934	16.3033	17.8674	20.0
2.5579	0.91	55000	2.4251	20.0544	7.8966	16.3889	17.9491	20.0
2.5242	0.99	60000	2.4097	20.1093	8.0572	16.4935	17.9823	20.0

Framework versions

• Transformers 4.23.0.dev0
• Pytorch 1.12.0
• Datasets 2.4.0
• Tokenizers 0.12.1