Modelpublic

toanduz/PRIMERA-multinews-lora-finetuned

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes4downloads

Model Card

PRIMERA-multinews-lora-finetuned

This model is a fine-tuned version of allenai/PRIMERA-multinews on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 1.6767
—Rouge1: 13.1661
—Rouge2: 6.075
—Rougel: 11.1948
—Rougelsum: 12.1382
—Gen Len: 20.0

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: 2e-05
—trainbatchsize: 2
—evalbatchsize: 2
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: linear
—num_epochs: 16
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
2.2385	1.0	4725	1.9183	15.8308	7.6039	12.6624	14.1866	20.0
2.1419	2.0	9450	1.8933	14.8545	6.8893	12.144	13.4623	20.0
2.1286	3.0	14175	1.8619	16.2585	8.1431	13.4226	14.8653	20.0
2.0669	4.0	18900	1.8129	15.9624	7.5293	13.3809	14.7765	20.0
2.0448	5.0	23625	1.7636	16.2801	8.045	13.6178	14.996	20.0
1.9831	6.0	28350	1.7037	13.8735	5.9956	10.8251	12.2545	20.0
1.9926	7.0	33075	1.7623	13.9591	5.8861	11.112	12.4349	20.0
1.99	8.0	37800	1.7247	13.1441	5.2565	10.7117	11.851	20.0
1.9495	9.0	42525	1.7065	12.4863	4.6444	10.0155	11.3874	20.0
1.9782	10.0	47250	1.6919	11.8394	4.0068	9.4554	10.6421	20.0
1.9087	11.0	51975	1.6910	13.011	5.5644	10.7255	11.8532	20.0
1.9693	12.0	56700	1.6872	13.2678	5.7966	11.0537	12.1103	20.0
1.9445	13.0	61425	1.7084	13.2757	5.9337	11.084	12.2354	20.0
1.9467	14.0	66150	1.6729	12.9202	5.424	10.5315	11.6661	20.0
1.9582	15.0	70875	1.6786	13.2851	6.0806	11.291	12.2518	20.0
1.9186	16.0	75600	1.6767	13.1661	6.075	11.1948	12.1382	20.0

Framework versions

—PEFT 0.12.0
—Transformers 4.43.2
—Pytorch 2.2.1+cu121
—Datasets 2.20.0
—Tokenizers 0.19.1