xlsr-big-kbnnn

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Wer: 0.0555

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: 0.0004
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 132
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1412	1.3652	200	0.9308	0.8207
0.6085	2.7304	400	0.1016	0.1634
0.1952	4.0956	600	0.0382	0.0936
0.0912	5.4608	800	0.0131	0.0637
0.0745	6.8259	1000	0.0106	0.0599
0.0537	8.1911	1200	0.0123	0.0605
0.0503	9.5563	1400	0.0087	0.0585
0.0457	10.9215	1600	0.0088	0.0569
0.0356	12.2867	1800	0.0102	0.0579
0.0328	13.6519	2000	0.0060	0.0589
0.0313	15.0171	2200	0.0032	0.0561
0.0313	16.3823	2400	0.0036	0.0581
0.0256	17.7474	2600	0.0047	0.0563
0.0234	19.1126	2800	0.0034	0.0516
0.0186	20.4778	3000	0.0029	0.0541
0.0205	21.8430	3200	0.0009	0.0527
0.0161	23.2082	3400	0.0045	0.0531
0.019	24.5734	3600	0.0021	0.0533
0.0162	25.9386	3800	0.0014	0.0518
0.0151	27.3038	4000	0.0046	0.0573
0.0162	28.6689	4200	0.0178	0.0555
0.0202	30.0341	4400	0.0040	0.0561
0.0204	31.3993	4600	0.0005	0.0529
0.018	32.7645	4800	0.0010	0.0561
0.0108	34.1297	5000	0.0001	0.0559
0.0124	35.4949	5200	0.0007	0.0625
0.011	36.8601	5400	0.0004	0.0523
0.0119	38.2253	5600	0.0019	0.0675
0.0138	39.5904	5800	0.0003	0.0585
0.0114	40.9556	6000	0.0011	0.0533
0.0094	42.3208	6200	0.0006	0.0543
0.0097	43.6860	6400	0.0002	0.0539
0.0079	45.0512	6600	0.0003	0.0523
0.0107	46.4164	6800	0.0001	0.0575
0.0077	47.7816	7000	0.0016	0.0527
0.008	49.1468	7200	0.0044	0.0603
0.0087	50.5119	7400	0.0001	0.0525
0.008	51.8771	7600	0.0001	0.0514
0.0064	53.2423	7800	0.0002	0.0579
0.0057	54.6075	8000	0.0003	0.0525
0.0061	55.9727	8200	0.0027	0.0547
0.0082	57.3379	8400	0.0004	0.0557
0.0065	58.7031	8600	0.0002	0.0555
0.0074	60.0683	8800	0.0000	0.0591
0.0066	61.4334	9000	0.0000	0.0543
0.0056	62.7986	9200	0.0000	0.0531
0.0043	64.1638	9400	0.0003	0.0549
0.0035	65.5290	9600	0.0000	0.0569
0.0049	66.8942	9800	0.0012	0.0603
0.0045	68.2594	10000	0.0001	0.0543
0.0037	69.6246	10200	0.0000	0.0537
0.0061	70.9898	10400	0.0000	0.0571
0.0041	72.3549	10600	0.0006	0.0569
0.0031	73.7201	10800	0.0000	0.0583
0.0042	75.0853	11000	0.0000	0.0561
0.0031	76.4505	11200	0.0000	0.0551
0.0027	77.8157	11400	0.0000	0.0567
0.0029	79.1809	11600	0.0000	0.0537
0.0026	80.5461	11800	0.0000	0.0539
0.001	81.9113	12000	0.0000	0.0529
0.0044	83.2765	12200	0.0000	0.0520
0.0016	84.6416	12400	0.0000	0.0559
0.0022	86.0068	12600	0.0000	0.0553
0.0018	87.3720	12800	0.0000	0.0557
0.0016	88.7372	13000	0.0000	0.0555
0.0012	90.1024	13200	0.0000	0.0561
0.001	91.4676	13400	0.0000	0.0575
0.0011	92.8328	13600	0.0000	0.0563
0.0008	94.1980	13800	0.0000	0.0569
0.001	95.5631	14000	0.0000	0.0561
0.0007	96.9283	14200	0.0000	0.0559
0.0009	98.2935	14400	0.0000	0.0553
0.0009	99.6587	14600	0.0000	0.0555

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1