ikema-asr-youtube-aug

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: 4.0169
• Cer: 0.5439

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: 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: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
11.3718	0.2188	100	3.7224	0.9906
3.6044	0.4376	200	3.6799	0.9909
3.5365	0.6565	300	3.7114	0.9908
3.2721	0.8753	400	3.5040	0.9440
2.202	1.0941	500	2.6818	0.7443
1.3265	1.3129	600	2.7658	0.6453
0.9665	1.5317	700	2.5355	0.6076
0.7187	1.7505	800	2.7831	0.6322
0.5584	1.9694	900	2.8070	0.6103
0.4578	2.1882	1000	3.0448	0.5877
0.4056	2.4070	1100	2.9589	0.5973
0.3737	2.6258	1200	2.6101	0.5738
0.3582	2.8446	1300	3.0568	0.5934
0.3437	3.0635	1400	2.9337	0.5579
0.2958	3.2823	1500	3.1510	0.5791
0.2859	3.5011	1600	3.0342	0.5787
0.2766	3.7199	1700	3.0843	0.5809
0.2666	3.9387	1800	2.9355	0.5636
0.2577	4.1575	1900	2.9394	0.5934
0.2392	4.3764	2000	3.0028	0.5690
0.2307	4.5952	2100	3.2121	0.5897
0.2182	4.8140	2200	2.8466	0.5560
0.2472	5.0328	2300	3.0391	0.5893
0.1969	5.2516	2400	3.0167	0.5617
0.2136	5.4705	2500	3.2553	0.5867
0.1955	5.6893	2600	3.0892	0.5480
0.1844	5.9081	2700	3.1589	0.5574
0.1858	6.1269	2800	3.1052	0.5507
0.1656	6.3457	2900	3.1258	0.5597
0.1661	6.5646	3000	3.1924	0.5619
0.1716	6.7834	3100	3.5032	0.6322
0.1852	7.0022	3200	2.8788	0.5526
0.1547	7.2210	3300	3.2716	0.5573
0.1614	7.4398	3400	3.7549	0.6075
0.1424	7.6586	3500	3.7023	0.5681
0.1366	7.8775	3600	3.4016	0.5658
0.1183	8.0963	3700	3.7288	0.5863
0.1393	8.3151	3800	3.5373	0.5679
0.1403	8.5339	3900	3.6973	0.5592
0.1338	8.7527	4000	3.5407	0.5701
0.1221	8.9716	4100	3.8244	0.5708
0.1211	9.1904	4200	3.2298	0.5473
0.1178	9.4092	4300	3.1331	0.5390
0.1199	9.6280	4400	3.1808	0.5474
0.1192	9.8468	4500	3.3093	0.5699
0.1158	10.0656	4600	3.5926	0.5676
0.111	10.2845	4700	3.4394	0.5476
0.1006	10.5033	4800	3.5034	0.5591
0.1087	10.7221	4900	3.4475	0.5506
0.1034	10.9409	5000	3.8657	0.5614
0.1069	11.1597	5100	3.7824	0.5949
0.0961	11.3786	5200	3.5943	0.5644
0.0902	11.5974	5300	3.4558	0.5715
0.0993	11.8162	5400	3.4136	0.5654
0.1084	12.0350	5500	3.3599	0.5634
0.0916	12.2538	5600	3.6072	0.5683
0.094	12.4726	5700	3.3698	0.5655
0.0867	12.6915	5800	3.4165	0.5610
0.0744	12.9103	5900	3.7541	0.5742
0.0972	13.1291	6000	3.4058	0.5690
0.0835	13.3479	6100	3.5592	0.5582
0.087	13.5667	6200	3.5826	0.5484
0.0786	13.7856	6300	3.4395	0.5639
0.0888	14.0044	6400	3.2782	0.5433
0.0697	14.2232	6500	3.9069	0.5629
0.0713	14.4420	6600	3.5987	0.5629
0.0732	14.6608	6700	3.5931	0.5581
0.0703	14.8796	6800	4.1842	0.6007
0.0642	15.0985	6900	3.6846	0.5515
0.0667	15.3173	7000	3.7597	0.5670
0.071	15.5361	7100	3.3775	0.5609
0.0722	15.7549	7200	3.5182	0.5545
0.0709	15.9737	7300	3.5706	0.5429
0.0639	16.1926	7400	3.8510	0.5562
0.0661	16.4114	7500	3.6689	0.5493
0.0623	16.6302	7600	3.7224	0.5498
0.0542	16.8490	7700	3.4815	0.5519
0.0615	17.0678	7800	3.5955	0.5511
0.0511	17.2867	7900	3.5042	0.5373
0.0559	17.5055	8000	3.6237	0.5545
0.0574	17.7243	8100	3.7682	0.5747
0.0553	17.9431	8200	3.8012	0.5583
0.0549	18.1619	8300	3.8540	0.5507
0.0555	18.3807	8400	3.7402	0.5787
0.0492	18.5996	8500	3.9611	0.5815
0.0513	18.8184	8600	3.8181	0.5694
0.0606	19.0372	8700	3.8336	0.5848
0.0487	19.2560	8800	4.1520	0.6026
0.0432	19.4748	8900	4.0908	0.5942
0.0426	19.6937	9000	3.8979	0.5654
0.0461	19.9125	9100	3.6559	0.5569
0.0526	20.1313	9200	3.4724	0.5381
0.047	20.3501	9300	3.8027	0.5639
0.0549	20.5689	9400	3.8417	0.5435
0.0396	20.7877	9500	3.9192	0.5718
0.0407	21.0066	9600	4.0680	0.5789
0.0393	21.2254	9700	4.1327	0.5871
0.0368	21.4442	9800	3.8152	0.5516
0.0461	21.6630	9900	3.9048	0.5679
0.0432	21.8818	10000	3.8295	0.5749
0.0316	22.1007	10100	4.0871	0.5639
0.0375	22.3195	10200	3.8720	0.5534
0.0391	22.5383	10300	3.7463	0.5469
0.0324	22.7571	10400	3.7395	0.5477
0.0311	22.9759	10500	3.7983	0.5411
0.0342	23.1947	10600	3.8577	0.5551
0.0335	23.4136	10700	3.7687	0.5498
0.0332	23.6324	10800	4.0054	0.5480
0.0316	23.8512	10900	3.9865	0.5626
0.035	24.0700	11000	3.9826	0.5504
0.0305	24.2888	11100	3.8950	0.5573
0.0313	24.5077	11200	3.7649	0.5590
0.0327	24.7265	11300	3.6834	0.5497
0.0273	24.9453	11400	3.6726	0.5503
0.0279	25.1641	11500	3.6909	0.5480
0.0286	25.3829	11600	3.8002	0.5439
0.0246	25.6018	11700	4.0034	0.5597
0.0286	25.8206	11800	3.8345	0.5543
0.0263	26.0394	11900	3.9427	0.5557
0.0233	26.2582	12000	4.0726	0.5433
0.0252	26.4770	12100	4.1124	0.5555
0.0219	26.6958	12200	3.8680	0.5427
0.0227	26.9147	12300	3.9605	0.5362
0.0226	27.1335	12400	3.8697	0.5456
0.019	27.3523	12500	4.0662	0.5467
0.0204	27.5711	12600	4.0304	0.5443
0.0212	27.7899	12700	3.9678	0.5442
0.0208	28.0088	12800	3.9714	0.5421
0.0206	28.2276	12900	3.9737	0.5403
0.0196	28.4464	13000	4.0019	0.5423
0.0191	28.6652	13100	4.0598	0.5471
0.0223	28.8840	13200	4.0701	0.5449
0.0151	29.1028	13300	4.0333	0.5432
0.0194	29.3217	13400	4.0175	0.5424
0.0188	29.5405	13500	4.0117	0.5449
0.0189	29.7593	13600	4.0113	0.5455
0.021	29.9781	13700	4.0165	0.5441

Framework versions

• Transformers 4.51.2
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1