base1

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

• Loss: 0.2996
• Cer: 0.1149

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: 0.0001
• train_batch_size: 32
• eval_batch_size: 8
• 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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
10.6595	0.8	100	14.7038	0.9746
6.0052	1.6	200	4.7152	0.9746
3.7094	2.4	300	3.6087	0.9746
3.3185	3.2	400	3.4963	0.9746
3.2647	4.0	500	3.3655	0.7936
3.1367	4.8	600	3.1738	0.7899
2.7489	5.6	700	2.2306	0.6644
1.8455	6.4	800	0.9514	0.2306
1.0656	7.2	900	0.5763	0.1712
0.7766	8.0	1000	0.4218	0.1481
0.6282	8.8	1100	0.3556	0.1352
0.5584	9.6	1200	0.3430	0.1313
0.4898	10.4	1300	0.3302	0.1291
0.4736	11.2	1400	0.3161	0.1240
0.4336	12.0	1500	0.3091	0.1257
0.4094	12.8	1600	0.2978	0.1215
0.4014	13.6	1700	0.2920	0.1209
0.3674	14.4	1800	0.3032	0.1238
0.3585	15.2	1900	0.2886	0.1200
0.3389	16.0	2000	0.2977	0.1206
0.3238	16.8	2100	0.2898	0.1172
0.314	17.6	2200	0.2835	0.1161
0.3135	18.4	2300	0.2934	0.1173
0.2958	19.2	2400	0.2912	0.1165
0.296	20.0	2500	0.2982	0.1167
0.2826	20.8	2600	0.2886	0.1167
0.2672	21.6	2700	0.2889	0.1165
0.2678	22.4	2800	0.2920	0.1163
0.2668	23.2	2900	0.2952	0.1162
0.2552	24.0	3000	0.2957	0.1157
0.2397	24.8	3100	0.2972	0.1151
0.2462	25.6	3200	0.2980	0.1146
0.2435	26.4	3300	0.2937	0.1140
0.2337	27.2	3400	0.3015	0.1153
0.246	28.0	3500	0.2996	0.1147
0.2389	28.8	3600	0.2984	0.1142
0.224	29.6	3700	0.2996	0.1149

Framework versions

• Transformers 4.17.0
• Pytorch 2.4.0
• Datasets 1.18.3
• Tokenizers 0.20.3