HuBERT_Jibbali_lang

This model is a fine-tuned version of facebook/hubert-large-ll60k on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2017
• Wer: 0.1944
• Cet: 0.1189

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cet
10.6563	0.99	56	5.6577	1.0	0.9812
3.3895	2.0	113	3.2018	1.0	0.9812
3.1588	2.99	169	3.1347	1.0	0.9812
3.1308	4.0	226	3.0567	1.0	0.9812
2.8933	4.99	282	2.8226	1.0	0.9353
2.5444	6.0	339	2.0947	1.0	0.8588
0.995	6.99	395	0.5049	0.4974	0.1654
0.3567	8.0	452	0.2622	0.2485	0.1132
0.2914	8.99	508	0.1980	0.2105	0.0749
0.14	10.0	565	0.2154	0.2069	0.0821
0.1442	10.99	621	0.1965	0.1988	0.0969
0.1401	12.0	678	0.2135	0.1937	0.0960
0.1019	12.99	734	0.2185	0.1948	0.1094
0.1088	14.0	791	0.1957	0.1966	0.1121
0.1314	14.99	847	0.1983	0.1933	0.1019
0.0522	16.0	904	0.2026	0.1944	0.1258
0.126	16.99	960	0.2033	0.1944	0.1142
0.1028	18.0	1017	0.1940	0.1974	0.1158
0.0767	18.99	1073	0.1969	0.1948	0.1149
0.0468	19.82	1120	0.2017	0.1944	0.1189

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu121
• Datasets 2.17.0
• Tokenizers 0.15.2