xlsr-syntesized-turkish-4-hour-llr

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: 3.7455
• Wer: 1.0

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: 1e-05
• 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
9.2067	0.52	100	7.7166	1.0
4.7666	1.04	200	4.5059	1.0
3.6173	1.56	300	3.6445	1.0
3.3747	2.08	400	3.4531	1.0
3.2351	2.6	500	3.3077	1.0
3.16	3.12	600	3.1965	1.0
3.092	3.65	700	3.1473	1.0
3.0051	4.17	800	3.1395	1.0
2.785	4.69	900	3.2007	1.0
2.3783	5.21	1000	3.1456	1.0
2.0074	5.73	1100	3.0829	1.0
1.767	6.25	1200	3.0563	1.0
1.4704	6.77	1300	3.0522	1.0
1.2749	7.29	1400	3.1219	1.0
1.12	7.81	1500	3.1256	1.0
1.0151	8.33	1600	3.1973	1.0
0.9098	8.85	1700	3.2901	1.0
0.8451	9.38	1800	3.2970	1.0
0.793	9.9	1900	3.3592	1.0
0.7518	10.42	2000	3.4237	1.0
0.6994	10.94	2100	3.4695	1.0
0.6686	11.46	2200	3.5052	1.0
0.6558	11.98	2300	3.5431	1.0
0.6247	12.5	2400	3.6077	1.0
0.615	13.02	2500	3.6611	1.0
0.5889	13.54	2600	3.5684	1.0
0.5702	14.06	2700	3.5891	1.0
0.5552	14.58	2800	3.6027	1.0
0.5367	15.1	2900	3.6398	1.0
0.5322	15.62	3000	3.6480	1.0
0.5051	16.15	3100	3.6087	1.0
0.5063	16.67	3200	3.6777	1.0
0.4869	17.19	3300	3.6848	1.0
0.4945	17.71	3400	3.6886	1.0
0.4848	18.23	3500	3.7217	1.0
0.489	18.75	3600	3.7360	1.0
0.4811	19.27	3700	3.7442	1.0
0.4665	19.79	3800	3.7455	1.0

Framework versions

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