wav2vec2-xls-r-300m-FL-af-1hr-v3

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.0397
• Wer: 0.4997
• Cer: 0.1532

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
10.726	2.6316	100	4.8546	1.0	1.0
3.892	5.2632	200	3.2785	1.0	1.0
3.0342	7.8947	300	2.8899	1.0	1.0
2.8665	10.5263	400	2.8735	1.0	1.0
2.8427	13.1579	500	2.8842	1.0	1.0
2.8293	15.7895	600	2.8202	1.0	0.9911
2.2253	18.4211	700	1.4009	0.8946	0.4027
0.7951	21.0526	800	0.9228	0.6411	0.2026
0.3831	23.6842	900	0.9069	0.5647	0.1782
0.2542	26.3158	1000	0.9096	0.5582	0.1730
0.1801	28.9474	1100	0.9426	0.5682	0.1725
0.1446	31.5789	1200	0.9889	0.5349	0.1666
0.1238	34.2105	1300	0.9938	0.5356	0.1644
0.1068	36.8421	1400	1.0244	0.5294	0.1647
0.1019	39.4737	1500	1.0253	0.5151	0.1609
0.0823	42.1053	1600	1.0374	0.5195	0.1612
0.0769	44.7368	1700	1.0574	0.5174	0.1600
0.0694	47.3684	1800	1.0425	0.5119	0.1569
0.0672	50.0	1900	1.0317	0.5054	0.1555
0.0563	52.6316	2000	1.0554	0.5036	0.1551
0.0541	55.2632	2100	1.0385	0.4992	0.1529
0.0505	57.8947	2200	1.0424	0.5003	0.1532

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1