es_xh

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5893
• Wer: 0.5947
• Cer: 0.4759

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
13.2606	0.4	500	3.9328	1.0	0.9985
3.2356	0.8	1000	1.7868	1.1103	0.7647
1.3553	1.2	1500	0.9949	0.9400	0.6160
1.0288	1.6	2000	0.8121	0.8587	0.5799
0.912	2.0	2500	0.7674	0.8885	0.5624
0.7449	2.4	3000	0.7171	0.8139	0.5611
0.6755	2.8	3500	0.6477	0.7770	0.5350
0.5974	3.2	4000	0.6217	0.7566	0.5268
0.5474	3.6	4500	0.6131	0.7395	0.5221
0.5241	4.0	5000	0.6161	0.7428	0.5269
0.4447	4.4	5500	0.5790	0.6977	0.5110
0.434	4.8	6000	0.5889	0.7103	0.5119
0.3912	5.2	6500	0.5804	0.6916	0.5095
0.3572	5.6	7000	0.5728	0.6997	0.5099
0.3546	6.0	7500	0.5640	0.6845	0.5070
0.2939	6.4	8000	0.5703	0.6677	0.4988
0.2884	6.8	8500	0.5500	0.6618	0.5000
0.2636	7.2	9000	0.5548	0.6534	0.4963
0.246	7.6	9500	0.5447	0.6437	0.4945
0.2358	8.0	10000	0.5480	0.6485	0.4945
0.2012	8.4	10500	0.5476	0.6447	0.4921
0.2038	8.8	11000	0.5590	0.6481	0.4935
0.1833	9.2	11500	0.5820	0.6435	0.4921
0.1695	9.6	12000	0.5649	0.6294	0.4902
0.1649	10.0	12500	0.5663	0.6255	0.4874
0.1382	10.4	13000	0.5676	0.6361	0.4874
0.1387	10.8	13500	0.5940	0.6251	0.4868
0.1252	11.2	14000	0.5833	0.6262	0.4848
0.1146	11.6	14500	0.5669	0.6223	0.4832
0.1119	12.0	15000	0.5697	0.6170	0.4835
0.0968	12.4	15500	0.5893	0.6096	0.4805
0.0929	12.8	16000	0.5790	0.6048	0.4793
0.0891	13.2	16500	0.5916	0.6018	0.4786
0.0784	13.6	17000	0.5921	0.6008	0.4771
0.079	14.0	17500	0.5918	0.5952	0.4761
0.07	14.4	18000	0.5930	0.5944	0.4761
0.0688	14.8	18500	0.5893	0.5947	0.4759

Framework versions

• Transformers 4.36.2
• Pytorch 1.13.0+cu116
• Datasets 2.15.0
• Tokenizers 0.15.2