wav2vec2-large-xls-r-300m-lg-cv-10hr-v3

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: 0.6399
• Wer: 0.5490
• Cer: 0.1258

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	0.9948	95	5.7544	1.0	1.0
11.4782	2.0	191	3.4141	1.0	1.0
3.8877	2.9948	286	2.9705	1.0	1.0
3.0666	4.0	382	2.8116	1.0	1.0
2.8721	4.9948	477	0.9460	0.9262	0.2276
1.6147	6.0	573	0.6163	0.8134	0.1855
0.6412	6.9948	668	0.4726	0.6816	0.1425
0.4424	8.0	764	0.4475	0.6449	0.1306
0.3408	8.9948	859	0.4403	0.6429	0.1310
0.2786	10.0	955	0.4409	0.6139	0.1252
0.24	10.9948	1050	0.4206	0.5878	0.1218
0.2111	12.0	1146	0.4501	0.5916	0.1194
0.1881	12.9948	1241	0.4514	0.5645	0.1140
0.1672	14.0	1337	0.4553	0.5761	0.1224
0.1532	14.9948	1432	0.4780	0.5764	0.1179
0.1421	16.0	1528	0.4795	0.5767	0.1177
0.1357	16.9948	1623	0.4573	0.5643	0.1189
0.1248	18.0	1719	0.4774	0.5679	0.1202
0.1176	18.9948	1814	0.5095	0.5659	0.1186
0.111	20.0	1910	0.4775	0.5562	0.1138
0.1093	20.9948	2005	0.5052	0.5465	0.1115
0.1017	22.0	2101	0.5074	0.5464	0.1123
0.1017	22.9948	2196	0.5003	0.5419	0.1135
0.0965	24.0	2292	0.5247	0.5420	0.1130
0.0947	24.9948	2387	0.5224	0.5474	0.1152
0.0903	26.0	2483	0.5124	0.5250	0.1089
0.0865	26.9948	2578	0.5339	0.5387	0.1100
0.0837	28.0	2674	0.5362	0.5340	0.1128
0.0836	28.9948	2769	0.5354	0.5276	0.1095
0.0773	30.0	2865	0.5512	0.5352	0.1101
0.075	30.9948	2960	0.5162	0.5102	0.1058
0.0723	32.0	3056	0.5296	0.5236	0.1057
0.0764	32.9948	3151	0.5447	0.5289	0.1083
0.0706	34.0	3247	0.5291	0.5355	0.1138
0.0694	34.9948	3342	0.5314	0.5244	0.1116
0.0679	36.0	3438	0.5199	0.5215	0.1135
0.0645	36.9948	3533	0.5555	0.5244	0.1118
0.0623	38.0	3629	0.5392	0.5266	0.1141
0.0622	38.9948	3724	0.5500	0.5248	0.1125
0.06	40.0	3820	0.5467	0.5197	0.1121
0.0598	40.9948	3915	0.5405	0.5161	0.1120

Framework versions

• Transformers 4.42.3
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1