wav2vec2-large-xls-r-300m-lg-cv-130hr-v1

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: 0.4019
• Wer: 0.2092
• Cer: 0.0456

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
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.7325	1.0	5194	0.2760	0.3441	0.0731
0.2056	2.0	10388	0.2507	0.2835	0.0629
0.1744	3.0	15582	0.2414	0.2721	0.0605
0.155	4.0	20776	0.2359	0.2618	0.0584
0.143	5.0	25970	0.2354	0.2577	0.0575
0.131	6.0	31164	0.2400	0.2551	0.0570
0.1208	7.0	36358	0.2460	0.2482	0.0555
0.1102	8.0	41552	0.2553	0.2439	0.0548
0.1001	9.0	46746	0.2441	0.2455	0.0547
0.0898	10.0	51940	0.2463	0.2423	0.0543
0.0795	11.0	57134	0.2577	0.2400	0.0528
0.0701	12.0	62328	0.2677	0.2374	0.0522
0.0609	13.0	67522	0.2741	0.2405	0.0527
0.0538	14.0	72716	0.2933	0.2396	0.0523
0.0471	15.0	77910	0.3096	0.2352	0.0517
0.0416	16.0	83104	0.3165	0.2311	0.0503
0.0374	17.0	88298	0.3294	0.2328	0.0505
0.0335	18.0	93492	0.3414	0.2325	0.0501
0.0301	19.0	98686	0.3379	0.2255	0.0487
0.0276	20.0	103880	0.3578	0.2220	0.0482
0.0253	21.0	109074	0.3701	0.2181	0.0476
0.0236	22.0	114268	0.3769	0.2181	0.0474
0.0217	23.0	119462	0.3808	0.2155	0.0470
0.0204	24.0	124656	0.3917	0.2124	0.0464
0.0193	25.0	129850	0.3963	0.2110	0.0459
0.0184	26.0	135044	0.3956	0.2111	0.0458
0.0174	27.0	140238	0.4046	0.2109	0.0459
0.0174	28.0	145432	0.3997	0.2096	0.0457
0.0169	29.0	150626	0.4014	0.2093	0.0456
0.0171	30.0	155820	0.4019	0.2092	0.0456

Framework versions

• Transformers 4.42.3
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1