wav2vec2-xls-r-phoneme-300m-tr-ogma-phoneme

This model is a fine-tuned version of patrickvonplaten/wav2vec2-xls-r-phoneme-300m-tr on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.6886
• Cer: 0.7624

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: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
9.8242	1.0	17	10.2075	1.7574
7.5139	2.0	34	6.2677	1.8317
4.93	3.0	51	4.4229	0.9950
3.787	4.0	68	3.6279	0.9604
3.2056	5.0	85	3.3155	0.9109
2.8302	6.0	102	3.0498	0.8515
2.6059	7.0	119	2.9567	0.8366
2.3369	8.0	136	2.8454	0.8465
2.0347	9.0	153	2.7595	0.8663
1.787	10.0	170	2.8327	0.8416
1.5493	11.0	187	2.7142	0.8465
1.3992	12.0	204	2.7668	0.8713
1.3539	13.0	221	2.7595	0.8465
1.1791	14.0	238	2.6278	0.8366
1.1649	15.0	255	2.8350	0.8564
1.0361	16.0	272	2.7286	0.7921
0.9179	17.0	289	2.6409	0.7772
0.8338	18.0	306	2.6040	0.7574
0.7847	19.0	323	2.7403	0.8564
0.82	20.0	340	2.6313	0.8168
0.753	21.0	357	2.5469	0.8168
0.6124	22.0	374	2.5799	0.7822
0.6236	23.0	391	2.6548	0.8069
0.5955	24.0	408	2.6331	0.8317
0.592	25.0	425	2.6168	0.8366
0.5169	26.0	442	2.6168	0.8069
0.5012	27.0	459	2.5482	0.7723
0.44	28.0	476	2.6088	0.8020
0.4243	29.0	493	2.6753	0.7871
0.4824	30.0	510	2.6886	0.7624

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2