wav2vec2M

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

• Loss: 2.3442
• Wer: 1

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: 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: 3000
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
13.3564	28.57	100	27.6137	1
12.2323	57.14	200	23.7860	1
6.6474	85.71	300	11.1610	1
3.5002	114.29	400	4.4376	1
3.2055	142.86	500	3.9892	1
3.1194	171.43	600	3.7170	1
3.0537	200.0	700	3.5853	1
3.009	228.57	800	3.4726	1
2.9618	257.14	900	3.3643	1
2.9248	285.71	1000	3.2705	1
2.8999	314.29	1100	3.2090	1
2.8771	342.86	1200	3.1526	1
2.8548	371.43	1300	3.1171	1
2.832	400.0	1400	3.0867	1
2.8173	428.57	1500	3.0383	1
2.7962	457.14	1600	3.0360	1
2.7794	485.71	1700	3.0202	1
2.7656	514.29	1800	3.0042	1
2.7539	542.86	1900	2.9969	1
2.741	571.43	2000	2.9984	1
2.7193	600.0	2100	2.9676	1
2.6152	628.57	2200	2.8485	1
2.335	657.14	2300	2.5736	1
1.9601	685.71	2400	2.3247	1
1.5744	714.29	2500	2.1688	1
1.306	742.86	2600	2.0462	1
1.0761	771.43	2700	2.0612	1
0.9187	800.0	2800	2.0175	1
0.8063	828.57	2900	2.0096	1
0.685	857.14	3000	2.0829	1
0.6094	885.71	3100	2.1365	1
0.5728	914.29	3200	2.1460	1
0.5295	942.86	3300	2.1939	1
0.4776	971.43	3400	2.1596	1
0.4391	1000.0	3500	2.1430	1
0.4192	1028.57	3600	2.2202	1
0.4191	1057.14	3700	2.2345	1
0.38	1085.71	3800	2.2531	1
0.3636	1114.29	3900	2.2553	1
0.3479	1142.86	4000	2.2299	1
0.3447	1171.43	4100	2.2707	1
0.3251	1200.0	4200	2.2855	1
0.3332	1228.57	4300	2.2808	1
0.3225	1257.14	4400	2.3391	1
0.3053	1285.71	4500	2.3565	1
0.3179	1314.29	4600	2.3616	1
0.2818	1342.86	4700	2.3429	1
0.2926	1371.43	4800	2.3507	1
0.2909	1400.0	4900	2.3501	1
0.2917	1428.57	5000	2.3442	1

Framework versions

• Transformers 4.28.0
• Pytorch 2.0.1+cu118
• Datasets 2.13.1
• Tokenizers 0.13.3