wav2vec2-base-ogma-phoneme

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

• Loss: nan
• Cer: 1.0

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: 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: 100
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
67.3394	1.0	5	62.1245	6.7376
73.2076	2.0	10	nan	6.7376
-257055208286820768.0000	3.0	15	nan	6.7376
64.2241	4.0	20	nan	6.7376
65.3601	5.0	25	62.1245	6.7376
64.2295	6.0	30	62.1157	6.7376
45.425	7.0	35	62.1251	6.7376
50.6118	8.0	40	nan	6.7178
64.6394	9.0	45	62.0582	6.6188
48.7615	10.0	50	nan	6.6188
54.5817	11.0	55	nan	6.4950
48.1198	12.0	60	nan	6.4950
56.9202	13.0	65	nan	6.3465
57.3656	14.0	70	nan	6.4406
68.163	15.0	75	61.7497	6.2129
56.806	16.0	80	nan	6.2129
69.1218	17.0	85	61.6119	5.7574
55.5282	18.0	90	61.5413	5.4158
-6752.4055	19.0	95	61.2303	4.9257
64.744	20.0	100	60.9641	4.4455
66.7382	21.0	105	60.3274	3.5198
-21060.9078	22.0	110	nan	3.5198
51.2619	23.0	115	59.9896	3.1089
51.398	24.0	120	nan	2.7772
63.6242	25.0	125	59.3321	2.5149
59.6308	26.0	130	58.7697	2.1931
62.0615	27.0	135	nan	1.8366
-46.2037	28.0	140	57.9474	1.7475
60.5632	29.0	145	57.5041	1.5941
55.4431	30.0	150	56.7507	1.4307
40.8661	31.0	155	56.6063	1.4059
63.784	32.0	160	56.1097	1.2327
42.2708	33.0	165	nan	1.2327
53.7813	34.0	170	nan	1.2426
57.459	35.0	175	55.8894	1.2228
58.9998	36.0	180	nan	1.0
0.0	37.0	185	nan	1.0
0.0	38.0	190	nan	1.0
0.0	39.0	195	nan	1.0
0.0	40.0	200	nan	1.0
0.0	41.0	205	nan	1.0
0.0	42.0	210	nan	1.0
0.0	43.0	215	nan	1.0
0.0	44.0	220	nan	1.0
0.0	45.0	225	nan	1.0
0.0	46.0	230	nan	1.0
0.0	47.0	235	nan	1.0
0.0	48.0	240	nan	1.0
0.0	49.0	245	nan	1.0
0.0	50.0	250	nan	1.0

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2