test3

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: 0.3231
• Cer: 0.1147

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
6.3356	0.8	100	3.4064	0.9746
3.2888	1.6	200	3.2514	0.8294
2.6425	2.4	300	1.3055	0.4934
1.155	3.2	400	0.4461	0.1792
0.7239	4.0	500	0.3535	0.1457
0.5787	4.8	600	0.3242	0.1313
0.5327	5.6	700	0.3335	0.1292
0.4861	6.4	800	0.3236	0.1278
0.4744	7.2	900	0.2971	0.1248
0.4373	8.0	1000	0.2832	0.1222
0.4102	8.8	1100	0.2920	0.1187
0.4044	9.6	1200	0.2860	0.1190
0.3591	10.4	1300	0.2963	0.1189
0.3572	11.2	1400	0.2879	0.1167
0.3428	12.0	1500	0.2887	0.1153
0.3227	12.8	1600	0.2808	0.1167
0.329	13.6	1700	0.2867	0.1159
0.2987	14.4	1800	0.2992	0.1174
0.2901	15.2	1900	0.2822	0.1143
0.2745	16.0	2000	0.3051	0.1157
0.2645	16.8	2100	0.2879	0.1132
0.2586	17.6	2200	0.2840	0.1140
0.2617	18.4	2300	0.3001	0.1148
0.2435	19.2	2400	0.3051	0.1161
0.2492	20.0	2500	0.3071	0.1152
0.2323	20.8	2600	0.2973	0.1135
0.2177	21.6	2700	0.3030	0.1151
0.218	22.4	2800	0.3076	0.1139
0.2173	23.2	2900	0.3205	0.1152
0.2105	24.0	3000	0.3139	0.1152
0.194	24.8	3100	0.3184	0.1145
0.2021	25.6	3200	0.3207	0.1148
0.1986	26.4	3300	0.3097	0.1135
0.1881	27.2	3400	0.3214	0.1142
0.2	28.0	3500	0.3204	0.1139
0.1906	28.8	3600	0.3237	0.1149
0.1753	29.6	3700	0.3231	0.1147

Framework versions

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