wav2vec2-1b-E30_speed

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: 0.5758
• Cer: 14.7028

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
16.6326	0.2580	200	4.7434	100.0
4.6388	0.5160	400	3.9414	88.5926
2.026	0.7741	600	1.7935	41.2711
1.1754	1.0321	800	1.2455	29.8990
0.8337	1.2901	1000	1.0037	25.8047
0.7399	1.5481	1200	0.9414	23.3846
0.6535	1.8062	1400	0.8053	21.4344
0.544	2.0642	1600	0.9218	23.1673
0.4264	2.3222	1800	0.8181	20.7354
0.387	2.5802	2000	0.7663	19.5900
0.3623	2.8383	2200	0.7863	20.2420
0.3416	3.0963	2400	0.8668	22.5975
0.2731	3.3543	2600	0.7251	18.6501
0.2394	3.6123	2800	0.6185	16.2183
0.2255	3.8703	3000	0.5926	15.5839
0.1874	4.1284	3200	0.5889	15.3900
0.1549	4.3864	3400	0.5765	15.1375
0.1514	4.6444	3600	0.5856	14.8614
0.1424	4.9024	3800	0.5758	14.7028

Framework versions

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