wav2vec2-1b-Yfreq_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: 0.8611
• Cer: 22.6210

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
14.3278	0.2580	200	3.6015	77.1323
2.1201	0.5160	400	1.8341	43.0627
1.4637	0.7741	600	1.6054	37.5587
1.2057	1.0321	800	1.3425	34.6805
0.9711	1.2901	1000	1.4205	35.2091
0.9025	1.5481	1200	1.3218	34.0872
0.8081	1.8062	1400	1.1956	30.2514
0.708	2.0642	1600	1.1235	28.9532
0.6006	2.3222	1800	1.2789	32.7655
0.5513	2.5802	2000	1.0509	27.1205
0.5326	2.8383	2200	1.1183	29.8990
0.4737	3.0963	2400	1.0453	26.8738
0.3829	3.3543	2600	0.9891	25.7460
0.3463	3.6123	2800	0.9118	23.8898
0.3313	3.8703	3000	0.9157	23.9074
0.2857	4.1284	3200	0.9297	24.4185
0.2432	4.3864	3400	0.8564	22.3273
0.2314	4.6444	3600	0.8760	22.5505
0.2125	4.9024	3800	0.8611	22.6210

Framework versions

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