agafurov/results

Evenki Wav2Vec2-large-960h

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the Evenki dataset. It achieves the following results on the evaluation set:

• Loss: 0.9874
• Wer: 100.0

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: 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: 250
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
15.7212	0.1595	50	15.5310	100.0
2.6269	0.3190	100	2.8415	99.7180
2.3072	0.4785	150	1.9787	99.4047
1.0452	0.6380	200	0.8638	100.0
1.1478	0.7974	250	2.7045	97.8850
1.3178	0.9569	300	0.8292	97.8850
1.3027	1.1164	350	6.3059	100.0
0.8389	1.2759	400	7.4609	100.0
2.0418	1.4354	450	6.3700	100.0
1.1457	1.5949	500	2.8782	100.0
0.787	1.7544	550	2.3695	100.0
0.6559	1.9139	600	3.4332	100.0
0.9913	2.0734	650	2.2318	100.0
0.7569	2.2329	700	2.1465	100.0
0.6979	2.3923	750	1.0902	100.0
1.2206	2.5518	800	2.2949	100.0
0.8054	2.7113	850	2.7187	100.0
0.686	2.8708	900	2.4085	100.0
0.8234	3.0303	950	0.9381	100.0
0.9062	3.1898	1000	1.0240	100.0
1.2424	3.3493	1050	1.1234	100.0
0.8654	3.5088	1100	4.7173	100.0
0.7106	3.6683	1150	0.9443	100.0
0.7094	3.8278	1200	0.9422	100.0
0.7053	3.9872	1250	1.1659	100.0
0.7629	4.1467	1300	1.1088	100.0
0.715	4.3062	1350	1.0223	100.0
1.2843	4.4657	1400	1.0170	100.0
0.6653	4.6252	1450	0.9866	100.0
0.668	4.7847	1500	0.9866	100.0
0.7227	4.9442	1550	0.9874	100.0

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1