README.md

metadata

tags:
  - generated_from_trainer
metrics:
  - wer
model-index:
  - name: tun_msa_wav2vec3
    results: []

tun_msa_wav2vec3

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5827
• Wer: 0.5757
• Cer: 0.1836

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: 1e-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 700
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
5.268	2.9221	900	2.2839	1.0005	0.5919
2.0283	5.8442	1800	0.9714	0.6901	0.2320
1.4551	8.7662	2700	0.7691	0.6608	0.2154
1.2854	11.6883	3600	0.7028	0.6369	0.2057
1.1381	14.6104	4500	0.6529	0.6172	0.1991
1.1017	17.5325	5400	0.6325	0.6050	0.1952
1.0674	20.4545	6300	0.6189	0.5958	0.1914
0.9982	23.3766	7200	0.6089	0.5918	0.1895
0.9585	26.2987	8100	0.5986	0.5860	0.1877
0.9073	29.2208	9000	0.5948	0.5822	0.1866
0.91	32.1429	9900	0.5915	0.5804	0.1854
0.8775	35.0649	10800	0.5885	0.5787	0.1849
0.8973	37.9870	11700	0.5877	0.5775	0.1844
0.8908	40.9091	12600	0.5857	0.5763	0.1841
0.8503	43.8312	13500	0.5831	0.5764	0.1837
0.8843	46.7532	14400	0.5831	0.5765	0.1838
0.8554	49.6753	15300	0.5827	0.5757	0.1836

Framework versions

• Transformers 4.41.2
• Pytorch 2.1.2
• Datasets 2.19.2
• Tokenizers 0.19.1