README.md

metadata

license: apache-2.0
tags:
  - automatic-speech-recognition
  - /workspace/data/uk/noizy_student_1/
  - generated_from_trainer
model-index:
  - name: ''
    results: []

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the /WORKSPACE/DATA/UK/NOIZY_STUDENT_1/ - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.1285
• Wer: 0.1821
• Cer: 0.0342

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
1.2323	3.22	500	0.0797	0.2816	0.4133
0.9826	6.45	1000	0.0514	0.1970	0.2688
0.8628	9.67	1500	0.0474	0.1649	0.2485
0.8348	12.9	2000	0.0467	0.1605	0.2460
0.8186	16.13	2500	0.0469	0.1608	0.2469
0.8011	19.35	3000	0.1620	0.2412	0.0468
0.807	22.58	3500	0.1737	0.2524	0.0498
0.7758	25.8	4000	0.1709	0.2536	0.0498
0.7923	29.03	4500	0.1645	0.2436	0.0474
0.7717	32.26	5000	0.1811	0.2636	0.0524
0.7447	35.48	5500	0.1635	0.2405	0.0468
0.7267	38.71	6000	0.1578	0.2354	0.0462
0.7046	41.93	6500	0.1555	0.2296	0.0444
0.6896	45.16	7000	0.1548	0.2272	0.0439
0.6575	48.38	7500	0.1432	0.2096	0.0399
0.6264	51.61	8000	0.1466	0.2056	0.0398
0.589	54.83	8500	0.1351	0.1943	0.0371
0.573	58.06	9000	0.1387	0.1934	0.0365
0.5537	61.29	9500	0.1328	0.1883	0.0353
0.544	64.51	10000	0.1285	0.1821	0.0342

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2
• Datasets 1.18.4.dev0
• Tokenizers 0.11.0