wav2vec2-large-xls-r-300m-dm32

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7826
• Accuracy: 0.7292

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 22
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.1930	34	0.6558	0.625
No log	2.3860	68	0.6781	0.6042
No log	3.5789	102	0.7024	0.4167
0.7018	4.7719	136	0.6712	0.5833
0.7018	5.9649	170	0.6506	0.625
0.7018	7.1579	204	0.6987	0.4375
0.6854	8.3509	238	0.6089	0.6875
0.6854	9.5439	272	0.5723	0.75
0.6854	10.7368	306	0.5788	0.7083
0.609	11.9298	340	0.6704	0.7083
0.609	13.1228	374	0.6395	0.7292
0.609	14.3158	408	0.6201	0.7292
0.4355	15.5088	442	0.7633	0.75
0.4355	16.7018	476	0.7693	0.7083
0.4355	17.8947	510	0.7872	0.6875
0.4355	19.0877	544	0.7747	0.7292
0.3369	20.2807	578	0.7819	0.7292
0.3369	21.4737	612	0.7826	0.7292

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1