excalibur12/k2e_asr-scr_w2v2-xls-r-300m_001

nia037-k2e_asr-scoring_wav2vec2-xls-r-300m_001

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: 1.3285
• Per: 0.1553
• Pcc: 0.5941
• Ctc Loss: 0.4571
• Mse Loss: 0.8435

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: 1
• seed: 1111
• gradient_accumulation_steps: 4
• 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: 750
• training_steps: 7500

Training results

Training Loss	Epoch	Step	Validation Loss	Per	Pcc	Ctc Loss	Mse Loss
16.8126	1.01	750	4.5828	0.9890	0.4831	3.7272	1.0554
4.3091	2.02	1500	3.9355	0.9627	0.5450	3.5438	0.8425
3.8448	3.02	2250	3.0227	0.7172	0.5510	2.5638	0.8462
2.2291	4.03	3000	1.6022	0.2220	0.5692	0.6960	0.8338
1.6362	5.04	3750	1.4686	0.1785	0.5888	0.5407	0.8311
1.4014	6.05	4500	1.3709	0.1676	0.6016	0.4993	0.7934
1.1905	7.06	5250	1.3089	0.1609	0.5990	0.4759	0.7807
1.0064	8.06	6000	1.3754	0.1574	0.5915	0.4659	0.8560
0.8526	9.07	6750	1.3020	0.1563	0.6048	0.4602	0.8179
0.7319	10.08	7500	1.3285	0.1553	0.5941	0.4571	0.8435

Framework versions

• Transformers 4.38.1
• Pytorch 2.0.1
• Datasets 2.16.1
• Tokenizers 0.15.2