test4

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: 0.3206
• Cer: 0.1143

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
7.0599	0.8	100	3.4931	0.9746
3.342	1.6	200	3.2933	0.9522
3.0383	2.4	300	2.0728	0.6779
1.489	3.2	400	0.5363	0.1846
0.7584	4.0	500	0.3476	0.1456
0.5789	4.8	600	0.3140	0.1339
0.5332	5.6	700	0.3108	0.1461
0.4893	6.4	800	0.3093	0.1251
0.4755	7.2	900	0.2992	0.1230
0.4362	8.0	1000	0.2870	0.1211
0.4088	8.8	1100	0.2939	0.1200
0.4038	9.6	1200	0.2801	0.1197
0.3571	10.4	1300	0.2996	0.1177
0.3583	11.2	1400	0.2820	0.1170
0.3449	12.0	1500	0.2858	0.1180
0.3215	12.8	1600	0.2790	0.1155
0.3249	13.6	1700	0.2802	0.1132
0.2993	14.4	1800	0.3010	0.1168
0.2903	15.2	1900	0.2809	0.1139
0.2763	16.0	2000	0.3017	0.1168
0.2647	16.8	2100	0.2906	0.1145
0.2587	17.6	2200	0.2811	0.1118
0.2589	18.4	2300	0.2920	0.1157
0.2409	19.2	2400	0.3025	0.1151
0.2468	20.0	2500	0.3079	0.1154
0.232	20.8	2600	0.2987	0.1135
0.2146	21.6	2700	0.3045	0.1164
0.2192	22.4	2800	0.3078	0.1153
0.2176	23.2	2900	0.3229	0.1157
0.2102	24.0	3000	0.3146	0.1151
0.1955	24.8	3100	0.3113	0.1144
0.2009	25.6	3200	0.3135	0.1146
0.1981	26.4	3300	0.3067	0.1132
0.1885	27.2	3400	0.3199	0.1150
0.1984	28.0	3500	0.3175	0.1149
0.1887	28.8	3600	0.3195	0.1147
0.1723	29.6	3700	0.3206	0.1143

Framework versions

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