korean_kws2

This model is a fine-tuned version of Kkonjeong/wav2vec2-base-korean on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3679
• Accuracy: 1.0

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	2	1.8970	0.3684
No log	2.0	4	1.8970	0.3684
No log	3.0	6	1.8769	0.1579
No log	4.0	8	1.8911	0.2105
No log	5.0	10	1.6754	0.4737
No log	6.0	12	1.6831	0.4211
No log	7.0	14	1.5253	0.5263
No log	8.0	16	1.3955	0.6842
No log	9.0	18	1.3499	0.7368
No log	10.0	20	1.2641	0.8947
No log	11.0	22	1.1479	0.8947
No log	12.0	24	1.0503	0.8947
No log	13.0	26	0.9568	0.8947
No log	14.0	28	0.8583	0.9474
No log	15.0	30	0.8090	1.0
No log	16.0	32	0.7293	1.0
No log	17.0	34	0.6670	1.0
No log	18.0	36	0.6095	1.0
No log	19.0	38	0.5649	1.0
No log	20.0	40	0.5258	1.0
No log	21.0	42	0.4889	1.0
No log	22.0	44	0.4592	1.0
No log	23.0	46	0.4349	1.0
No log	24.0	48	0.4166	1.0
No log	25.0	50	0.4028	1.0
No log	26.0	52	0.3908	1.0
No log	27.0	54	0.3812	1.0
No log	28.0	56	0.3745	1.0
No log	29.0	58	0.3698	1.0
No log	30.0	60	0.3679	1.0

Framework versions

• Transformers 5.15.0
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2