wav2vec2-xlsr-hsd-binary-malayalam

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: 0.0081
• Accuracy: 0.9888
• F1: 0.9887

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: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.6188	1.0	25	0.5817	0.7191	0.6774
0.4606	2.0	50	0.4867	0.7528	0.7456
0.2219	3.0	75	0.1340	0.9663	0.9662
0.0807	4.0	100	0.0736	0.9888	0.9887
0.0444	5.0	125	0.0710	0.9775	0.9774
0.038	6.0	150	0.0660	0.9888	0.9887
0.0295	7.0	175	0.0640	0.9888	0.9887
0.0165	8.0	200	0.0674	0.9888	0.9887
0.0041	9.0	225	0.0002	1.0	1.0
0.0336	10.0	250	0.0354	0.9888	0.9887
0.0044	11.0	275	0.0622	0.9888	0.9887
0.0094	12.0	300	0.0901	0.9888	0.9887
0.0177	13.0	325	0.0870	0.9775	0.9773
0.0036	14.0	350	0.0415	0.9888	0.9887
0.0116	15.0	375	0.0874	0.9888	0.9887
0.0006	16.0	400	0.0675	0.9888	0.9887
0.0229	17.0	425	0.0328	0.9888	0.9887
0.0021	18.0	450	0.0144	0.9888	0.9887
0.0005	19.0	475	0.0103	0.9888	0.9887
0.0059	20.0	500	0.0081	0.9888	0.9887

Framework versions

• Transformers 4.53.3
• Pytorch 2.9.1+cu128
• Datasets 4.4.1
• Tokenizers 0.21.2