ssc-bew-model

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.2698
• Cer: 0.4050
• Wer: 0.9687

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
4.8123	0.4484	100	3.0133	0.9969	1.0
3.1265	0.8969	200	2.9396	0.9969	1.0
3.119	1.3453	300	2.8454	0.9969	1.0
3.0907	1.7937	400	2.8643	0.9969	1.0
3.0756	2.2422	500	2.9632	0.9969	1.0
2.9941	2.6906	600	2.8723	0.9969	1.0
2.8902	3.1390	700	2.6961	0.9892	1.0
2.7428	3.5874	800	2.4343	0.8631	0.9999
2.5463	4.0359	900	2.3476	0.7009	1.0
2.3062	4.4843	1000	2.1183	0.6930	1.0
2.2073	4.9327	1100	2.0210	0.7117	1.0
2.0765	5.3812	1200	2.0935	0.6305	0.9960
2.031	5.8296	1300	1.9194	0.6221	0.9994
1.9854	6.2780	1400	1.8179	0.5811	0.9999
1.8907	6.7265	1500	1.7837	0.5585	0.9998
1.8584	7.1749	1600	1.8593	0.5447	0.9931
1.7839	7.6233	1700	1.8705	0.5141	1.0417
1.785	8.0717	1800	1.7395	0.5033	0.9858
1.6759	8.5202	1900	1.7391	0.4723	0.9718
1.7218	8.9686	2000	1.7175	0.4940	0.9723
1.609	9.4170	2100	1.7110	0.4781	0.9772
1.5811	9.8655	2200	1.7124	0.4658	0.9843
1.5396	10.3139	2300	1.6364	0.4590	0.9560
1.5133	10.7623	2400	1.6455	0.4580	0.9590
1.4639	11.2108	2500	1.7604	0.4770	0.9661
1.4046	11.6592	2600	1.7846	0.4839	1.0466
1.3616	12.1076	2700	1.6502	0.4380	0.9452
1.3035	12.5561	2800	1.6649	0.4458	0.9521
1.3431	13.0045	2900	1.6487	0.4386	0.9449
1.2072	13.4529	3000	1.6000	0.4358	0.9402
1.1978	13.9013	3100	1.6768	0.4380	0.9613
1.1108	14.3498	3200	1.7206	0.4302	0.9412
1.1131	14.7982	3300	1.6861	0.4315	0.9715
1.0874	15.2466	3400	1.5880	0.4253	0.9327
1.0333	15.6951	3500	1.5706	0.4154	0.9273
0.9514	16.1435	3600	1.7136	0.4228	0.9538
0.9313	16.5919	3700	1.8036	0.4289	1.0255
0.9902	17.0404	3800	1.6053	0.4120	0.9242
0.8452	17.4888	3900	1.5672	0.4104	0.9327
0.872	17.9372	4000	1.6529	0.4131	0.9536
0.7782	18.3857	4100	1.8549	0.4384	1.0661
0.7984	18.8341	4200	1.8437	0.4153	1.0265
0.7615	19.2825	4300	1.7319	0.4127	0.9696
0.7246	19.7309	4400	1.7560	0.4035	0.9411
0.698	20.1794	4500	1.8200	0.4254	0.9628
0.6601	20.6278	4600	1.8046	0.4220	0.9428
0.6534	21.0762	4700	1.9306	0.4108	0.9498
0.6144	21.5247	4800	1.8637	0.4024	0.9563
0.6418	21.9731	4900	1.9459	0.4060	0.9565
0.5814	22.4215	5000	1.9776	0.4344	1.0098
0.577	22.8700	5100	2.0336	0.4109	0.9729
0.5315	23.3184	5200	2.1168	0.4078	0.9863
0.5363	23.7668	5300	2.0074	0.4066	0.9515
0.5649	24.2152	5400	2.0267	0.4149	0.9734
0.5181	24.6637	5500	1.9906	0.4076	0.9469
0.4848	25.1121	5600	2.1771	0.4014	0.9471
0.4804	25.5605	5700	2.0896	0.3972	0.9382
0.5048	26.0090	5800	2.1047	0.3946	0.9295
0.4581	26.4574	5900	2.1378	0.4016	0.9482
0.4616	26.9058	6000	2.1853	0.4071	0.9466
0.4075	27.3543	6100	2.2196	0.3982	0.9473
0.4338	27.8027	6200	2.1815	0.4027	0.9511
0.4475	28.2511	6300	2.2522	0.4019	0.9564
0.4115	28.6996	6400	2.2593	0.4024	0.9580
0.3882	29.1480	6500	2.2502	0.4072	0.9666
0.4021	29.5964	6600	2.2698	0.4050	0.9687

Framework versions

• Transformers 4.57.2
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.0