wav2vec2-large-xls-r-300m-irish-colab_test

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

• Loss: 1.7839
• Wer: 0.6220

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 90
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
10.0428	2.94	50	4.1311	1.0
3.2917	5.88	100	3.1468	1.0
3.0221	8.82	150	2.9848	1.0
2.9795	11.76	200	2.9567	1.0
2.9379	14.71	250	2.9463	1.0
2.9068	17.65	300	2.8330	1.0
2.5088	20.59	350	1.9807	0.9535
1.6188	23.53	400	1.4254	0.8398
1.0435	26.47	450	1.3668	0.7807
0.7212	29.41	500	1.3914	0.7476
0.5456	32.35	550	1.5495	0.7470
0.4297	35.29	600	1.4751	0.6960
0.3533	38.24	650	1.5157	0.6909
0.2899	41.18	700	1.5394	0.6879
0.2529	44.12	750	1.6186	0.6903
0.2413	47.06	800	1.6386	0.6954
0.2113	50.0	850	1.6906	0.6778
0.1769	52.94	900	1.6918	0.6575
0.1622	55.88	950	1.7313	0.6572
0.1564	58.82	1000	1.7701	0.6510
0.1637	61.76	1050	1.6800	0.6444
0.148	64.71	1100	1.7306	0.6477
0.1385	67.65	1150	1.7605	0.6408
0.1264	70.59	1200	1.7534	0.6244
0.1157	73.53	1250	1.7906	0.6381
0.1027	76.47	1300	1.7803	0.6265
0.1061	79.41	1350	1.7617	0.6259
0.0934	82.35	1400	1.7649	0.6253
0.0904	85.29	1450	1.7713	0.6187
0.0911	88.24	1500	1.7839	0.6220

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.18.3
• Tokenizers 0.10.3