metadata

license: apache-2.0
tags:
  - generated_from_trainer
metrics:
  - wer
  - cer
model-index:
  - name: wav2vec2-xls-r-300m-th-cv11_0
    results: []
datasets:
  - mozilla-foundation/common_voice_11_0
language:
  - th
pipeline_tag: automatic-speech-recognition

wav2vec2-xls-r-300m-th-cv11_0

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.3391
Wer: 0.2915
Cer: 0.0651
Clean Cer: 0.0508
Learning Rate: 0.0000

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: 16
eval_batch_size: 16
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 32
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 10
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer	Clean Cer	Rate
7.5397	0.37	500	3.5716	1.0	0.9811	0.9774	0.0001
1.7478	0.75	1000	0.7702	0.8097	0.2296	0.1746	0.0001
0.7687	1.12	1500	0.4997	0.5392	0.1415	0.1182	0.0001
0.6064	1.5	2000	0.4270	0.4956	0.1238	0.1001	0.0001
0.5473	1.87	2500	0.3809	0.4489	0.1105	0.0898	0.0001
0.454	2.24	3000	0.3585	0.4256	0.1021	0.0813	0.0001
0.4219	2.62	3500	0.3375	0.4063	0.0974	0.0777	0.0001
0.4075	2.99	4000	0.3274	0.4036	0.0948	0.0746	0.0001
0.3355	3.37	4500	0.3257	0.3782	0.0898	0.0729	0.0001
0.3203	3.74	5000	0.3024	0.3561	0.0830	0.0659	0.0001
0.3151	4.11	5500	0.3038	0.3606	0.0830	0.0653	0.0001
0.2713	4.49	6000	0.3052	0.3595	0.0832	0.0655	0.0001
0.2685	4.86	6500	0.2933	0.3436	0.0796	0.0628	0.0001
0.2379	5.24	7000	0.3020	0.3362	0.0763	0.0608	0.0000
0.224	5.61	7500	0.2874	0.3265	0.0745	0.0589	0.0000
0.2204	5.98	8000	0.2922	0.3191	0.0724	0.0576	0.0000
0.1927	6.36	8500	0.3107	0.3163	0.0719	0.0568	0.0000
0.1875	6.73	9000	0.3034	0.3084	0.0703	0.0554	0.0000
0.1786	7.11	9500	0.3210	0.3107	0.0702	0.0553	0.0000
0.1606	7.48	10000	0.3231	0.3062	0.0688	0.0541	0.0000
0.1594	7.85	10500	0.3234	0.3033	0.0680	0.0535	0.0000
0.1498	8.23	11000	0.3276	0.3035	0.0680	0.0530	0.0000
0.1396	8.6	11500	0.3265	0.2975	0.0668	0.0520	0.0000
0.142	8.98	12000	0.3236	0.2930	0.0659	0.0515	0.0000
0.1242	9.35	12500	0.3403	0.2921	0.0655	0.0511	0.0000
0.1225	9.72	13000	0.3391	0.2915	0.0651	0.0508	0.0000

Framework versions

Transformers 4.27.0.dev0
Pytorch 1.13.1+cu116
Datasets 2.9.0
Tokenizers 0.13.2