led-large-16384-cnn_dailymail

This model is a fine-tuned version of allenai/led-base-16384 on the cnn_dailymail dataset. It achieves the following results on the evaluation set:

• Loss: 1.5544
• Rouge1: 0.3870
• Rouge2: 0.1736
• Rougel: 0.2599
• Rougelsum: 0.3653

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
1.9531	0.4	500	1.8639	0.3485	0.1441	0.2275	0.3288
1.9563	0.8	1000	1.8260	0.3538	0.1482	0.2315	0.3343
1.7176	1.2	1500	1.8208	0.3628	0.1527	0.2383	0.3433
1.7197	1.6	2000	1.8162	0.3696	0.1602	0.2434	0.3486
1.8086	2.0	2500	1.7924	0.3558	0.1533	0.2334	0.3361
1.2448	2.4	3000	1.8510	0.3703	0.1591	0.2447	0.3483
1.3574	2.8	3500	1.8277	0.3741	0.1593	0.2422	0.3540
1.0966	3.2	4000	1.8924	0.3682	0.1576	0.2424	0.3479
0.9938	3.6	4500	1.8957	0.3723	0.1599	0.2451	0.3511
1.0735	4.0	5000	1.8772	0.3653	0.1557	0.2399	0.3454
0.9106	4.4	5500	1.9401	0.3720	0.1585	0.2436	0.3504
1.015	4.8	6000	1.9320	0.3725	0.1570	0.2429	0.3515
1.7854	0.36	6500	1.7800	0.3624	0.1544	0.2390	0.3422
1.9079	0.39	7000	1.7629	0.3573	0.1553	0.2352	0.3370
1.7606	3.34	7500	1.6902	0.3783	0.1673	0.2521	0.3570
1.7571	3.57	8000	1.6563	0.3802	0.1691	0.2538	0.3587
1.6602	3.79	8500	1.6439	0.3814	0.1693	0.2548	0.3600
1.6614	4.01	9000	1.6312	0.3812	0.1691	0.2544	0.3599
1.668	4.24	9500	1.6189	0.3815	0.1689	0.2550	0.3603
1.6491	4.46	10000	1.6172	0.3799	0.1681	0.2540	0.3586
1.5994	4.68	10500	1.6132	0.3825	0.1702	0.2560	0.3610
1.6493	4.9	11000	1.6093	0.3828	0.1701	0.2561	0.3613
1.6769	5.13	11500	1.6074	0.3831	0.1706	0.2569	0.3619
1.6554	5.35	12000	1.6044	0.3817	0.1695	0.2559	0.3605
1.6155	5.57	12500	1.6010	0.3825	0.1700	0.2561	0.3608
1.5863	5.8	13000	1.5981	0.3829	0.1704	0.2569	0.3614
1.6306	6.02	13500	1.6004	0.3831	0.1702	0.2563	0.3618
1.6425	6.24	14000	1.5987	0.3821	0.1698	0.2561	0.3610
1.6863	6.46	14500	1.5876	0.3837	0.1710	0.2569	0.3622
1.6085	6.69	15000	1.5815	0.3836	0.1717	0.2573	0.3621
1.6267	6.91	15500	1.5792	0.3852	0.1722	0.2579	0.3633
1.5637	7.13	16000	1.5768	0.3830	0.1709	0.2568	0.3611
1.5586	7.36	16500	1.5740	0.3833	0.1706	0.2567	0.3617
1.5389	7.58	17000	1.5689	0.3858	0.1729	0.2590	0.3640
1.5694	7.8	17500	1.5645	0.3853	0.1731	0.2589	0.3636
1.5265	8.02	18000	1.5621	0.3871	0.1733	0.2596	0.3654
1.5273	8.25	18500	1.5624	0.3861	0.1726	0.2588	0.3646
1.5148	8.47	19000	1.5602	0.3866	0.1733	0.2592	0.3651
1.532	8.69	19500	1.5599	0.3859	0.1732	0.2593	0.3642
1.5113	8.92	20000	1.5602	0.3877	0.1748	0.2606	0.3658
1.5133	9.14	20500	1.5595	0.3855	0.1725	0.2587	0.3637
1.4875	9.36	21000	1.5572	0.3873	0.1741	0.2600	0.3654
1.5038	9.59	21500	1.5557	0.3860	0.1728	0.2590	0.3641
1.5062	9.81	22000	1.5544	0.3870	0.1736	0.2599	0.3653

Framework versions

• Transformers 4.27.1
• Pytorch 2.0.0+cu118
• Datasets 2.10.1
• Tokenizers 0.13.2