Wav2vec2-xls-r-phoneme-300m-tr

This model is a fine-tuned version of wav2vec2-xls-r-300m on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.6380
• PER: 0.1664

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: 0.0005
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	PER
13.6687	0.92	100	12.4567	1.0
3.4219	1.83	200	3.4704	1.0
3.1846	2.75	300	3.2281	0.9935
2.0076	3.67	400	1.7415	0.5222
1.0244	4.59	500	1.0290	0.3323
0.7095	5.5	600	0.8424	0.2859
0.619	6.42	700	0.7389	0.2232
0.3541	7.34	800	0.7049	0.2043
0.2946	8.26	900	0.7065	0.2153
0.2868	9.17	1000	0.6840	0.2115
0.2245	10.09	1100	0.6714	0.1952
0.1394	11.01	1200	0.6864	0.1954
0.1288	11.93	1300	0.6696	0.2017
0.1568	12.84	1400	0.6468	0.1843
0.1269	13.76	1500	0.6736	0.1965
0.1101	14.68	1600	0.6689	0.1915
0.1388	15.6	1700	0.6690	0.1782
0.0739	16.51	1800	0.6364	0.1734
0.0897	17.43	1900	0.6480	0.1748
0.0795	18.35	2000	0.6356	0.1695
0.0823	19.27	2100	0.6382	0.1685

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.8.1
• Datasets 1.16.2.dev0
• Tokenizers 0.10.3