wav2vec2-large-mms-1b-kyrgyz-colab

This model is a fine-tuned version of facebook/mms-1b-all on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 3.6994
• Wer: 1.0028

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-05
• train_batch_size: 2
• 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: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
8.657	0.59	1000	5.3559	1.0
3.7087	1.18	2000	4.2908	1.0
3.4394	1.76	3000	4.1226	1.0
3.3918	2.35	4000	4.0324	1.0
3.3357	2.94	5000	3.9328	1.0
3.2915	3.53	6000	3.8575	0.9998
3.245	4.12	7000	3.7375	1.0001
3.219	4.71	8000	3.7220	1.0003
3.1878	5.29	9000	3.7172	1.0008
3.1826	5.88	10000	3.7064	1.0012
3.1613	6.47	11000	3.7029	1.0014
3.1548	7.06	12000	3.6923	1.0011
3.1403	7.65	13000	3.7013	1.0010
3.1286	8.24	14000	3.6862	1.0017
3.1184	8.82	15000	3.7129	1.0028
3.1198	9.41	16000	3.7063	1.0022
3.1051	10.0	17000	3.7093	1.0018
3.098	10.59	18000	3.6952	1.0031
3.1001	11.18	19000	3.7042	1.0031
3.0984	11.76	20000	3.6972	1.0020
3.0873	12.35	21000	3.7018	1.0026
3.0852	12.94	22000	3.7005	1.0025
3.0902	13.53	23000	3.6972	1.0026
3.081	14.12	24000	3.6984	1.0024
3.0777	14.71	25000	3.6994	1.0028

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0