wav2vec2-large-mn-pretrain-42h-100-epochs

This model is a fine-tuned version of bayartsogt/wav2vec2-large-mn-pretrain-42h on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 6.4172
• Wer: 1.0
• Cer: 0.9841

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
7.6418	1.59	400	6.4239	1.0	0.9841
5.5936	3.19	800	6.4154	1.0	0.9841
5.5208	4.78	1200	6.5248	1.0	0.9841
5.4869	6.37	1600	6.3805	1.0	0.9841
5.4757	7.97	2000	6.3988	1.0	0.9841
5.4624	9.56	2400	6.4058	1.0	0.9841
5.517	11.16	2800	6.3991	1.0	0.9841
5.4821	12.75	3200	6.4066	1.0	0.9841
5.487	14.34	3600	6.4281	1.0	0.9841
5.4786	15.93	4000	6.4174	1.0	0.9841
5.5017	17.53	4400	6.4338	1.0	0.9841
5.4967	19.12	4800	6.4653	1.0	0.9841
5.4619	20.72	5200	6.4499	1.0	0.9841
5.4883	22.31	5600	6.4345	1.0	0.9841
5.4899	23.9	6000	6.4224	1.0	0.9841
5.493	25.5	6400	6.4374	1.0	0.9841
5.4549	27.09	6800	6.4320	1.0	0.9841
5.4531	28.68	7200	6.4137	1.0	0.9841
5.4738	30.28	7600	6.4155	1.0	0.9841
5.4309	31.87	8000	6.4193	1.0	0.9841
5.4669	33.47	8400	6.4109	1.0	0.9841
5.47	35.06	8800	6.4111	1.0	0.9841
5.4623	36.65	9200	6.4102	1.0	0.9841
5.4583	38.25	9600	6.4150	1.0	0.9841
5.4551	39.84	10000	6.4172	1.0	0.9841

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0
• Datasets 2.1.0
• Tokenizers 0.12.1