wav2vec2-base-ms-with-lm

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: 1.1256
• Wer: 0.3991

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: 4
• 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: 1000
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.0672	1.85	500	2.9666	1.0
1.6566	3.7	1000	0.7862	0.6177
0.6424	5.56	1500	0.6600	0.5478
0.4438	7.41	2000	0.6148	0.4835
0.3133	9.26	2500	0.6949	0.4516
0.2822	11.11	3000	0.6844	0.4533
0.2338	12.96	3500	0.7682	0.4634
0.2246	14.81	4000	0.8370	0.4925
0.1957	16.67	4500	0.8554	0.4377
0.1809	18.52	5000	0.7211	0.4505
0.1496	20.37	5500	0.8081	0.4354
0.1369	22.22	6000	0.9099	0.4360
0.1285	24.07	6500	0.8051	0.4231
0.1174	25.93	7000	0.9041	0.4505
0.1074	27.78	7500	0.8096	0.4310
0.0904	29.63	8000	0.8589	0.4237
0.0975	31.48	8500	0.9019	0.4142
0.0806	33.33	9000	0.8966	0.4382
0.0772	35.19	9500	1.0612	0.4181
0.0738	37.04	10000	0.8979	0.4215
0.0678	38.89	10500	0.9342	0.4103
0.0626	40.74	11000	0.9992	0.4187
0.0641	42.59	11500	1.0101	0.4120
0.0565	44.44	12000	0.9841	0.4287
0.0535	46.3	12500	1.0049	0.4097
0.0512	48.15	13000	0.9569	0.4080
0.0466	50.0	13500	0.9863	0.4276
0.0514	51.85	14000	0.9594	0.4120
0.0397	53.7	14500	0.9775	0.4103
0.0394	55.56	15000	1.0077	0.4131
0.0393	57.41	15500	0.9763	0.4108
0.0355	59.26	16000	1.1432	0.4226
0.0323	61.11	16500	1.1553	0.4192
0.0344	62.96	17000	1.0437	0.4203
0.0296	64.81	17500	1.0227	0.4209
0.0267	66.67	18000	1.0408	0.4287
0.0283	68.52	18500	1.0811	0.4192
0.0243	70.37	19000	1.0697	0.4075
0.0238	72.22	19500	1.1047	0.4164
0.0234	74.07	20000	1.1585	0.4142
0.02	75.93	20500	1.1183	0.4187
0.0231	77.78	21000	1.0923	0.4080
0.018	79.63	21500	1.1299	0.4058
0.0179	81.48	22000	1.0963	0.4008
0.0152	83.33	22500	1.0676	0.4002
0.0142	85.19	23000	1.1026	0.4047
0.0159	87.04	23500	1.0876	0.4058
0.0144	88.89	24000	1.0943	0.3963
0.0152	90.74	24500	1.0827	0.4075
0.0155	92.59	25000	1.0982	0.4019
0.0124	94.44	25500	1.1284	0.3985
0.0132	96.3	26000	1.1233	0.3991
0.0109	98.15	26500	1.1196	0.3980
0.0102	100.0	27000	1.1256	0.3991

Framework versions

• Transformers 4.24.0
• Pytorch 2.0.0+cu118
• Datasets 1.18.3
• Tokenizers 0.13.3