xlsr-syntesized-turkish-2-hour-hlr

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: 0.1675
• Wer: 0.1901

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
6.2039	1.04	100	3.4534	1.0
3.1581	2.08	200	3.0148	1.0
2.2036	3.12	300	0.9141	0.7416
0.6325	4.17	400	0.3548	0.3585
0.3945	5.21	500	0.2863	0.3056
0.294	6.25	600	0.2475	0.2906
0.2358	7.29	700	0.2092	0.2510
0.177	8.33	800	0.1983	0.2595
0.1476	9.38	900	0.1856	0.2249
0.1267	10.42	1000	0.1706	0.2153
0.1132	11.46	1100	0.1785	0.2320
0.0979	12.5	1200	0.1779	0.2010
0.0836	13.54	1300	0.1688	0.1968
0.0738	14.58	1400	0.1688	0.1948
0.068	15.62	1500	0.1678	0.1915
0.0636	16.67	1600	0.1683	0.1899
0.0604	17.71	1700	0.1660	0.1891
0.0573	18.75	1800	0.1699	0.1903
0.0548	19.79	1900	0.1675	0.1901

Framework versions

• Transformers 4.26.0
• Pytorch 2.0.1+cu118
• Datasets 2.9.0
• Tokenizers 0.13.3