anvitamanne/base-model

Base Model

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: 513.6357
• Wer: 0.4037
• Cer: 0.1660

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: AdamW with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1711.3334	0.86	1000	768.0388	0.8588	0.3197
672.7345	1.72	2000	522.8168	0.6091	0.2202
574.6395	2.58	3000	495.6673	0.5407	0.2048
518.2652	3.44	4000	472.2298	0.5068	0.1910
485.7279	4.3	5000	448.1584	0.4797	0.1835
456.6944	5.17	6000	459.5286	0.4703	0.1805
440.0209	6.03	7000	490.1409	0.4549	0.1780
424.1306	6.89	8000	458.7100	0.4455	0.1754
413.0438	7.75	9000	446.2839	0.4371	0.1735
382.4416	8.61	10000	499.2264	0.4338	0.1728
370.5859	9.47	11000	489.0332	0.4243	0.1692
352.6317	10.33	12000	491.2080	0.4133	0.1664
371.7963	11.19	13000	464.0348	0.4109	0.1657
343.1062	12.05	14000	488.7343	0.4134	0.1676
332.8081	12.91	15000	518.1512	0.4044	0.1649
323.9083	13.78	16000	507.0865	0.4072	0.1667
323.2671	14.64	17000	513.6357	0.4037	0.1660

Framework versions

• Transformers 4.38.2
• Pytorch 2.1.0+cu118
• Datasets 3.6.0
• Tokenizers 0.15.2