wav2vec2-1b-E10_speed_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: 0.7350
• Cer: 19.3492

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
12.3554	0.2580	200	4.7531	99.0543
2.9802	0.5160	400	2.5589	60.7495
1.3723	0.7741	600	1.8529	48.9074
1.0598	1.0321	800	1.2929	34.4749
0.861	1.2901	1000	1.2272	31.8903
0.736	1.5481	1200	1.0361	26.4333
0.646	1.8062	1400	0.9841	26.2864
0.5576	2.0642	1600	1.0104	26.6095
0.4699	2.3222	1800	1.0558	26.8680
0.4313	2.5802	2000	1.0000	25.7989
0.4008	2.8383	2200	0.9699	25.7754
0.3416	3.0963	2400	0.8346	22.4213
0.2761	3.3543	2600	0.7851	20.2303
0.2554	3.6123	2800	0.7723	20.0012
0.2318	3.8703	3000	0.8079	21.3228
0.2013	4.1284	3200	0.8115	21.1290
0.1714	4.3864	3400	0.7670	19.6840
0.158	4.6444	3600	0.7683	19.8896
0.1471	4.9024	3800	0.7350	19.3492

Framework versions

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