wav2vec2-1b-E50_pause

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5661
• Cer: 14.9025

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: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
9.6567	0.2580	200	4.5898	72.2862
1.9249	0.5160	400	1.6835	41.2594
1.1868	0.7741	600	1.2607	30.9857
0.9419	1.0321	800	0.9705	23.5021
0.7837	1.2901	1000	1.0440	25.7460
0.6973	1.5481	1200	0.8581	21.7105
0.6151	1.8062	1400	0.7932	20.0658
0.5503	2.0642	1600	0.9359	23.5844
0.4372	2.3222	1800	0.8297	21.2700
0.4198	2.5802	2000	0.7865	19.9307
0.3897	2.8383	2200	0.6945	18.1156
0.3544	3.0963	2400	0.6895	18.2096
0.338	3.3543	2600	0.6466	16.9290
0.301	3.6123	2800	0.7000	18.7030
0.276	3.8703	3000	0.5962	15.8717
0.2448	4.1284	3200	0.6127	16.2300
0.2117	4.3864	3400	0.6149	16.4591
0.2113	4.6444	3600	0.6135	16.1654
0.188	4.9024	3800	0.5661	14.9025

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1