wav2vec2-xls-r-300m-paper

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the HTS98/ORIGINAL_VER1.2 - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.7895
• Wer: 0.4398

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: 5e-05
• train_batch_size: 10
• 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: 420
• num_epochs: 50.0

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	1.0	335	3.7157	1.0
6.2976	2.0	670	3.3644	1.0
3.2342	3.0	1005	2.4597	0.9739
3.2342	4.0	1340	1.4160	0.7444
1.2813	5.0	1675	1.1338	0.6543
0.7279	6.0	2010	1.0020	0.5856
0.7279	7.0	2345	0.8435	0.4823
0.5226	8.0	2680	0.8757	0.5078
0.4218	9.0	3015	0.7895	0.4398
0.4218	10.0	3350	0.7992	0.4228
0.3421	11.0	3685	0.8118	0.4307
0.287	12.0	4020	0.8215	0.4248
0.287	13.0	4355	0.8603	0.4077
0.2415	14.0	4690	0.8329	0.3886
0.2132	15.0	5025	0.8728	0.3955
0.2132	16.0	5360	0.8741	0.3918
0.1857	17.0	5695	0.8633	0.3675
0.1673	18.0	6030	0.8884	0.3804
0.1673	19.0	6365	0.9141	0.3679
0.1479	20.0	6700	0.9568	0.3605
0.1386	21.0	7035	0.9341	0.3630
0.1386	22.0	7370	0.9645	0.3537
0.1233	23.0	7705	0.9729	0.3567
0.1177	24.0	8040	1.0013	0.3454
0.1177	25.0	8375	1.0323	0.3597
0.1061	26.0	8710	1.0269	0.3456
0.1028	27.0	9045	1.0042	0.3424
0.1028	28.0	9380	1.0424	0.3394
0.0961	29.0	9715	1.0600	0.3412
0.0949	30.0	10050	1.0512	0.3389
0.0949	31.0	10385	1.0957	0.3389
0.0878	32.0	10720	1.0924	0.3311
0.0852	33.0	11055	1.0859	0.3366
0.0852	34.0	11390	1.1498	0.3450
0.0837	35.0	11725	1.0844	0.3329
0.0814	36.0	12060	1.1051	0.3321
0.0814	37.0	12395	1.0878	0.3311
0.0793	38.0	12730	1.1377	0.3286
0.0759	39.0	13065	1.1136	0.3246
0.0759	40.0	13400	1.1216	0.3268
0.0726	41.0	13735	1.1300	0.3253
0.0715	42.0	14070	1.1507	0.3262
0.0715	43.0	14405	1.1562	0.3275
0.0711	44.0	14740	1.1486	0.3219
0.0699	45.0	15075	1.1580	0.3194
0.0699	46.0	15410	1.1580	0.3195
0.0667	47.0	15745	1.1504	0.3212
0.0667	48.0	16080	1.1580	0.3203
0.0667	49.0	16415	1.1698	0.3192
0.0664	50.0	16750	1.1744	0.3192

Framework versions

• Transformers 4.31.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.7.0
• Tokenizers 0.13.2