xlsr-nm-clp

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: 1.3632
• Wer: 0.5241

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: Use 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: 132
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.0552	4.8780	200	3.0646	1.0
3.0248	9.7561	400	2.9305	1.0
2.8381	14.6341	600	2.7349	1.0
2.2963	19.5122	800	1.9857	0.9550
1.3557	24.3902	1000	1.3196	0.7685
0.6411	29.2683	1200	1.3063	0.6881
0.394	34.1463	1400	1.2477	0.6527
0.2608	39.0244	1600	1.1584	0.6013
0.1804	43.9024	1800	1.2374	0.6013
0.1442	48.7805	2000	1.3478	0.5643
0.1264	53.6585	2200	1.2854	0.5740
0.0892	58.5366	2400	1.2293	0.5900
0.0813	63.4146	2600	1.2025	0.5482
0.0597	68.2927	2800	1.3339	0.5466
0.0495	73.1707	3000	1.4527	0.5595
0.0453	78.0488	3200	1.4188	0.5257
0.0402	82.9268	3400	1.2740	0.5289
0.0367	87.8049	3600	1.3237	0.5161
0.0324	92.6829	3800	1.3321	0.5177
0.0267	97.5610	4000	1.3632	0.5241

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.4.0
• Datasets 3.0.1
• Tokenizers 0.20.0