wav2vec-turkish-300m-xls-2

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5268
• Wer: 0.4094

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: 0.0003
• train_batch_size: 32
• 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: 0.1
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5798	0.29	400	1.0275	0.8828
0.8353	0.58	800	0.9331	0.8265
0.7665	0.88	1200	0.8985	0.8207
0.7054	1.17	1600	0.7412	0.7435
0.6731	1.46	2000	0.6856	0.7321
0.6646	1.75	2400	0.7127	0.7589
0.6425	2.05	2800	0.6633	0.7016
0.5738	2.34	3200	0.6461	0.6872
0.5851	2.63	3600	0.6337	0.6808
0.5839	2.92	4000	0.6459	0.6951
0.5458	3.21	4400	0.6234	0.6699
0.5359	3.51	4800	0.6429	0.6777
0.5408	3.8	5200	0.6547	0.6833
0.5169	4.09	5600	0.6038	0.6444
0.4805	4.38	6000	0.5888	0.6439
0.4892	4.67	6400	0.5840	0.6349
0.4795	4.97	6800	0.5705	0.6327
0.4497	5.26	7200	0.6103	0.6621
0.4506	5.55	7600	0.5813	0.6328
0.4513	5.84	8000	0.5776	0.6423
0.4254	6.14	8400	0.6039	0.6218
0.424	6.43	8800	0.6233	0.6208
0.4246	6.72	9200	0.5717	0.6248
0.4233	7.01	9600	0.5588	0.5968
0.3829	7.3	10000	0.5472	0.5922
0.397	7.6	10400	0.5176	0.5713
0.3813	7.89	10800	0.5004	0.5721
0.3623	8.18	11200	0.5643	0.5959
0.3551	8.47	11600	0.5771	0.5949
0.3685	8.77	12000	0.5878	0.6092
0.3562	9.06	12400	0.5197	0.5660
0.3275	9.35	12800	0.5242	0.5536
0.3378	9.64	13200	0.5141	0.5627
0.3476	9.93	13600	0.5140	0.5657
0.3272	10.23	14000	0.5235	0.5599
0.3152	10.52	14400	0.5018	0.5521
0.3119	10.81	14800	0.5034	0.5576
0.298	11.1	15200	0.5228	0.5649
0.2877	11.4	15600	0.5256	0.5592
0.2954	11.69	16000	0.5207	0.5513
0.2962	11.98	16400	0.4810	0.5348
0.2741	12.27	16800	0.4870	0.5278
0.2701	12.56	17200	0.4870	0.5366
0.2731	12.86	17600	0.4736	0.5274
0.2653	13.15	18000	0.4971	0.5340
0.252	13.44	18400	0.5104	0.5340
0.2579	13.73	18800	0.4838	0.5434
0.2457	14.02	19200	0.5106	0.5189
0.2403	14.32	19600	0.4655	0.5141
0.2335	14.61	20000	0.4887	0.5200
0.2414	14.9	20400	0.4792	0.5146
0.237	15.19	20800	0.4746	0.5063
0.2236	15.49	21200	0.4740	0.4985
0.2215	15.78	21600	0.4687	0.5031
0.2186	16.07	22000	0.5013	0.5106
0.2132	16.36	22400	0.4958	0.5095
0.2085	16.65	22800	0.4865	0.4841
0.2069	16.95	23200	0.4686	0.4924
0.1895	17.24	23600	0.4950	0.4862
0.2019	17.53	24000	0.4707	0.4875
0.1941	17.82	24400	0.4667	0.4838
0.1901	18.12	24800	0.4899	0.4846
0.179	18.41	25200	0.4840	0.4795
0.1817	18.7	25600	0.4878	0.4804
0.1771	18.99	26000	0.5001	0.4767
0.1674	19.28	26400	0.5023	0.4749
0.1716	19.58	26800	0.4710	0.4677
0.1723	19.87	27200	0.4855	0.4685
0.1647	20.16	27600	0.5059	0.4699
0.1557	20.45	28000	0.4829	0.4611
0.1621	20.75	28400	0.4833	0.4634
0.1543	21.04	28800	0.5226	0.4699
0.1463	21.33	29200	0.5186	0.4654
0.1513	21.62	29600	0.5028	0.4760
0.1474	21.91	30000	0.4965	0.4573
0.1405	22.21	30400	0.4657	0.4511
0.1385	22.5	30800	0.5062	0.4537
0.1351	22.79	31200	0.4843	0.4524
0.132	23.08	31600	0.4935	0.4484
0.1289	23.37	32000	0.5018	0.4491
0.1279	23.67	32400	0.4874	0.4432
0.1291	23.96	32800	0.4813	0.4404
0.1251	24.25	33200	0.4866	0.4412
0.1189	24.54	33600	0.4975	0.4420
0.1172	24.84	34000	0.4829	0.4363
0.1169	25.13	34400	0.4943	0.4333
0.1142	25.42	34800	0.5113	0.4326
0.1081	25.71	35200	0.5121	0.4314
0.1119	26.0	35600	0.5067	0.4316
0.1043	26.3	36000	0.5221	0.4292
0.1017	26.59	36400	0.5230	0.4264
0.1035	26.88	36800	0.5141	0.4267
0.0957	27.17	37200	0.5320	0.4231
0.0994	27.47	37600	0.5173	0.4180
0.0947	27.76	38000	0.5218	0.4162
0.0932	28.05	38400	0.5163	0.4181
0.0912	28.34	38800	0.5277	0.4151
0.0928	28.63	39200	0.5152	0.4136
0.0918	28.93	39600	0.5145	0.4125
0.0852	29.22	40000	0.5257	0.4108
0.0904	29.51	40400	0.5239	0.4092
0.0858	29.8	40800	0.5268	0.4094

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2