wav2vec2-large-xlsr-sw_ndizi_782_100_epochs

This model is a fine-tuned version of smutuvi/wav2vec2-large-xlsr-sw on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.1009
• Wer: 0.4847

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: 0.0003
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4035	4.79	400	1.2492	0.5608
0.8489	9.58	800	1.0208	0.5114
0.632	14.37	1200	1.3292	0.5306
0.4653	19.16	1600	1.5159	0.5109
0.3598	23.95	2000	1.4650	0.5450
0.2776	28.74	2400	1.8568	0.5124
0.218	33.53	2800	2.0913	0.5188
0.1711	38.32	3200	2.2706	0.5035
0.141	43.11	3600	2.3050	0.5094
0.1162	47.9	4000	2.4539	0.5025
0.1007	52.69	4400	2.4754	0.5020
0.0881	57.49	4800	2.5512	0.5030
0.0816	62.28	5200	2.6458	0.5064
0.0792	67.07	5600	2.7869	0.5025
0.06	71.86	6000	2.9063	0.5040
0.0594	76.65	6400	2.8363	0.5049
0.0527	81.44	6800	3.0801	0.4921
0.0473	86.23	7200	3.0959	0.4867
0.0471	91.02	7600	3.0942	0.4852
0.0405	95.81	8000	3.1009	0.4847

Framework versions

• Transformers 4.37.1
• Pytorch 2.2.1+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0