speller-t5-8

This model is a fine-tuned version of sberbank-ai/ruT5-base on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1626
• Rouge1: 14.5345
• Rouge2: 9.6847
• Rougel: 14.4144
• Rougelsum: 14.3544
• Gen Len: 37.7748

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum	Gen Len
0.9929	0.04	500	0.5644	12.8915	6.982	12.7499	12.6126	39.3604
1.2778	0.07	1000	0.4272	13.8323	8.1081	13.7538	13.6059	38.8288
0.6384	0.11	1500	0.3654	14.1341	8.5586	14.034	13.9139	38.4865
0.5993	0.15	2000	0.3318	14.009	8.3655	13.9339	13.7688	38.8018
0.5327	0.18	2500	0.2969	14.2342	8.7838	14.1141	13.9339	37.7928
0.4831	0.22	3000	0.2761	14.5345	9.6847	14.4144	14.3544	38.9369
0.4582	0.25	3500	0.2562	14.5345	9.6847	14.4144	14.3544	39.2162
0.3983	0.29	4000	0.2449	14.5345	9.6847	14.4144	14.3544	38.9459
0.4459	0.33	4500	0.2422	14.4717	9.5045	14.4144	14.2206	38.8378
0.4073	0.36	5000	0.2375	14.5345	9.6847	14.4144	14.3544	38.3964
0.4047	0.4	5500	0.2263	14.5345	9.6847	14.4144	14.3544	39.2703
0.3423	0.44	6000	0.2208	14.5345	9.6847	14.4144	14.3544	38.5405
0.3348	0.47	6500	0.2109	14.5345	9.6847	14.4144	14.3544	37.7568
0.3421	0.51	7000	0.2053	14.5345	9.6847	14.4144	14.3544	37.7117
0.3319	0.54	7500	0.2025	14.2342	8.7838	14.1141	14.0541	37.5586
0.3239	0.58	8000	0.1991	14.2342	8.7838	14.1141	14.0541	38.0541
0.2963	0.62	8500	0.1959	14.2342	8.7838	14.1141	14.0541	38.0
0.3117	0.65	9000	0.1899	14.2342	8.7838	14.1141	14.0541	37.7117
0.2737	0.69	9500	0.1898	14.2342	8.7838	14.1141	14.0541	37.5135
0.3425	0.73	10000	0.1830	14.2342	8.7838	14.1141	14.0541	37.6306
0.2986	0.76	10500	0.1831	14.5345	9.6847	14.4144	14.3544	37.7658
0.3312	0.8	11000	0.1734	14.2342	8.7838	14.1141	14.0541	37.973
0.3461	0.83	11500	0.1753	14.5345	9.6847	14.4144	14.3544	37.6847
0.2786	0.87	12000	0.1740	14.5345	9.6847	14.4144	14.3544	37.7748
0.2911	0.91	12500	0.1672	14.5345	9.6847	14.4144	14.3544	37.7387
0.2618	0.94	13000	0.1691	14.5345	9.6847	14.4144	14.3544	37.5135
0.2844	0.98	13500	0.1626	14.5345	9.6847	14.4144	14.3544	37.7748

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1+cu116
• Datasets 2.9.0
• Tokenizers 0.13.2