wav2vec2-large-xls-r-300m

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: 3.6299
• Cer: 69.5709

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: 8
• 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: 100
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
56.9189	0.13	200	17.3335	100.0
24.9477	0.26	400	5.2898	93.4002
22.4883	0.39	600	5.1855	91.7503
23.6577	0.52	800	5.5828	93.8097
12.0508	0.64	1000	5.5876	92.9788
6.9102	0.77	1200	5.2180	93.1153
5.2937	0.9	1400	4.6411	93.6613
4.6224	1.03	1600	4.7614	93.7088
4.6148	1.16	1800	4.8569	93.0916
4.6129	1.29	2000	4.6641	92.5693
4.5837	1.42	2200	4.7341	93.5664
4.5465	1.55	2400	4.7580	93.3646
4.5042	1.68	2600	4.8342	93.4180
4.4535	1.81	2800	4.8323	93.4952
4.3942	1.93	3000	4.7891	92.1538
4.347	2.06	3200	4.5461	92.4150
4.3064	2.19	3400	4.7049	92.0114
4.301	2.32	3600	4.6162	92.3438
4.2107	2.45	3800	4.5264	92.3497
4.1768	2.58	4000	4.4067	87.1090
4.1395	2.71	4200	4.5684	90.6048
4.0568	2.84	4400	4.4142	87.7856
4.0007	2.97	4600	4.2861	87.7619
3.9426	3.09	4800	4.2750	87.4236
3.8207	3.22	5000	4.2217	84.2424
3.7678	3.35	5200	4.1129	81.1324
3.7522	3.48	5400	4.0891	83.2512
3.6666	3.61	5600	4.0328	80.0107
3.587	3.74	5800	4.0241	80.2837
3.5184	3.87	6000	4.0116	81.8209
3.449	4.0	6200	3.9031	76.4437
3.3503	4.13	6400	3.8643	75.8680
3.2541	4.26	6600	3.8508	72.4850
3.1987	4.38	6800	3.8425	71.8203
3.1435	4.51	7000	3.7289	71.6719
3.0583	4.64	7200	3.6855	70.1703
3.0443	4.77	7400	3.6365	69.6540
3.0508	4.9	7600	3.6299	69.5709

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.0
• Tokenizers 0.15.0