wav2vec2-base-960h-finetuned-ks

This model is a fine-tuned version of facebook/wav2vec2-base-960h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.6449
• Accuracy: 0.1069

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: 128
• eval_batch_size: 128
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 1024
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.2
• lr_scheduler_warmup_steps: 10
• training_steps: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	1	2.6379	0.0840
1.3193	2.0	3	2.6377	0.0840
1.1536	3.0	4	2.6374	0.0763
0.8255	4.0	6	2.6377	0.0763
0.8247	5.0	8	2.6390	0.0763
0.8247	6.0	9	2.6387	0.0840
1.1536	7.0	11	2.6415	0.0992
1.3183	8.0	12	2.6408	0.0916
1.3183	9.0	13	2.6402	0.0992
1.3176	10.0	15	2.6414	0.0992
1.1517	11.0	16	2.6419	0.0992
0.823	12.0	18	2.6426	0.0992
0.8222	13.0	20	2.6449	0.1069
0.8222	14.0	21	2.6467	0.0992
1.1534	15.0	23	2.6469	0.0916
1.3186	16.0	24	2.6464	0.0840
1.3186	17.0	25	2.6460	0.0840
1.3143	18.0	27	2.6454	0.0916
1.1482	19.0	28	2.6450	0.0840
0.8229	20.0	30	2.6450	0.0840

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.0+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0