wav2vec2-xls-r-300m-scandinavian-E2-100h-30-epochs-20250129_v2.2

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

• Loss: 0.1222
• Wer: 21.6178
• Cer: 4.5303

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 5000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.3227	0.5831	1000	3.2779	99.9982	99.0647
1.3019	1.1662	2000	0.8990	74.2685	20.8956
0.5718	1.7493	3000	0.3181	43.9153	10.5261
0.4453	2.3324	4000	0.2424	36.6078	8.5666
0.3749	2.9155	5000	0.2031	34.0956	7.8005
0.3686	3.4985	6000	0.1904	32.1394	7.2849
0.3006	4.0816	7000	0.1812	30.3624	6.9172
0.2924	4.6647	8000	0.1689	29.4684	6.6876
0.2346	5.2478	9000	0.1550	28.0239	6.2367
0.2272	5.8309	10000	0.1484	27.8096	6.2244
0.2469	6.4140	11000	0.1536	27.1668	6.0145
0.2508	6.9971	12000	0.1471	26.8398	5.9329
0.2041	7.5802	13000	0.1448	26.1213	5.7387
0.1724	8.1633	14000	0.1412	25.7832	5.6786
0.1813	8.7464	15000	0.1389	25.7333	5.6383
0.1942	9.3294	16000	0.1373	25.4119	5.5902
0.171	9.9125	17000	0.1382	25.3584	5.5399
0.2056	10.4956	18000	0.1392	24.9649	5.5094
0.1462	11.0787	19000	0.1363	24.4846	5.3482
0.1669	11.6618	20000	0.1360	24.3756	5.3060
0.1471	12.2449	21000	0.1298	24.2630	5.2992
0.166	12.8280	22000	0.1321	24.2020	5.2576
0.1785	13.4111	23000	0.1285	24.1725	5.2435
0.1785	13.9942	24000	0.1251	23.8714	5.1578
0.1407	14.5773	25000	0.1287	23.9970	5.2290
0.1093	15.1603	26000	0.1303	23.5204	5.0653
0.1005	15.7434	27000	0.1295	23.3061	5.0284
0.1399	16.3265	28000	0.1283	23.3043	4.9784
0.1239	16.9096	29000	0.1277	23.0937	4.9328
0.1208	17.4927	30000	0.1277	23.1103	4.9606
0.1013	18.0758	31000	0.1248	22.7538	4.8388
0.0926	18.6589	32000	0.1207	22.5783	4.7880
0.1262	19.2420	33000	0.1246	22.5525	4.8188
0.1168	19.8251	34000	0.1254	22.3363	4.7420
0.1158	20.4082	35000	0.1242	22.2625	4.7180
0.1149	20.9913	36000	0.1210	22.1461	4.6841
0.0818	21.5743	37000	0.1223	22.0223	4.6709
0.0982	22.1574	38000	0.1237	22.0371	4.6625
0.105	22.7405	39000	0.1205	21.8191	4.5981
0.104	23.3236	40000	0.1208	21.8431	4.5917
0.101	23.9067	41000	0.1217	21.7970	4.5954
0.0714	24.4898	42000	0.1230	21.7637	4.5707
0.0753	25.0729	43000	0.1208	21.6732	4.5405
0.0744	25.6560	44000	0.1213	21.6935	4.5439
0.0954	26.2391	45000	0.1200	21.6861	4.5402
0.0909	26.8222	46000	0.1215	21.6658	4.5411
0.0848	27.4052	47000	0.1224	21.6399	4.5328
0.0851	27.9883	48000	0.1228	21.6104	4.5331
0.0689	28.5714	49000	0.1225	21.6547	4.5390
0.0855	29.1545	50000	0.1224	21.6178	4.5334
0.0857	29.7376	51000	0.1222	21.6178	4.5303

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0