README.md

metadata

library_name: transformers
language:
  - lg
license: mit
base_model: facebook/w2v-bert-2.0
tags:
  - generated_from_trainer
datasets:
  - yogera
metrics:
  - wer
model-index:
  - name: wav2vec2-bert
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Yogera
          type: yogera
        metrics:
          - name: Wer
            type: wer
            value: 0.2828398665554629

wav2vec2-bert

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the Yogera dataset. It achieves the following results on the evaluation set:

• Loss: 0.5797
• Wer: 0.2828
• Cer: 0.0600

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.9705	1.0	20	2.7517	0.9999	0.8635
1.6229	2.0	40	0.7456	0.7959	0.1852
0.5343	3.0	60	0.4631	0.5219	0.1081
0.3188	4.0	80	0.3815	0.3815	0.0788
0.2219	5.0	100	0.3961	0.3946	0.0786
0.18	6.0	120	0.3851	0.3340	0.0698
0.1388	7.0	140	0.4074	0.3508	0.0723
0.1262	8.0	160	0.3911	0.3110	0.0665
0.0883	9.0	180	0.4100	0.3058	0.0636
0.0681	10.0	200	0.4071	0.3096	0.0644
0.0553	11.0	220	0.4505	0.3132	0.0662
0.0546	12.0	240	0.4836	0.3156	0.0681
0.038	13.0	260	0.4465	0.2941	0.0631
0.0374	14.0	280	0.5048	0.3071	0.0645
0.0275	15.0	300	0.4964	0.2901	0.0600
0.0207	16.0	320	0.4497	0.2946	0.0611
0.0163	17.0	340	0.4992	0.2867	0.0599
0.0248	18.0	360	0.5160	0.2952	0.0626
0.0139	19.0	380	0.5196	0.2960	0.0612
0.0177	20.0	400	0.5026	0.2933	0.0608
0.013	21.0	420	0.5212	0.2852	0.0603
0.0103	22.0	440	0.5277	0.2847	0.0599
0.01	23.0	460	0.5850	0.2802	0.0597
0.0088	24.0	480	0.5362	0.2841	0.0584
0.0074	25.0	500	0.5401	0.2974	0.0612
0.0087	26.0	520	0.5586	0.2858	0.0593
0.0077	27.0	540	0.5924	0.2859	0.0614
0.0114	28.0	560	0.5722	0.2828	0.0599
0.0118	29.0	580	0.5475	0.2842	0.0579
0.0102	30.0	600	0.5249	0.2851	0.0584
0.0083	31.0	620	0.5724	0.2801	0.0584
0.0087	32.0	640	0.5742	0.2799	0.0596
0.0098	33.0	660	0.5797	0.2828	0.0600

Framework versions

• Transformers 4.45.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.1
• Tokenizers 0.20.1