wav2vec2_xls_r_300m_BIG-C_Bemba_100hr_v1

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: 0.4735
• Model Preparation Time: 0.005
• Wer: 0.4226
• Cer: 0.1075

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: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
1.4689	1.0	1549	inf	0.005	0.4875	0.1241
0.5854	2.0	3098	inf	0.005	0.4906	0.1216
0.5464	3.0	4647	inf	0.005	0.4519	0.1163
0.5213	4.0	6196	inf	0.005	0.4321	0.1118
0.4994	5.0	7745	inf	0.005	0.4350	0.1131
0.4831	6.0	9294	inf	0.005	0.4462	0.1256
0.4959	7.0	10843	inf	0.005	0.7363	0.1926
2.8986	8.0	12392	nan	0.005	0.7915	0.2142
0.0205	9.0	13941	nan	0.005	0.7915	0.2142
0.3227	10.0	15490	nan	0.005	0.7915	0.2142
0.4651	11.0	17039	nan	0.005	0.7915	0.2142
0.0061	12.0	18588	nan	0.005	0.7915	0.2142
0.1837	13.0	20137	nan	0.005	0.7915	0.2142
0.0034	14.0	21686	nan	0.005	0.7915	0.2142
0.0246	15.0	23235	nan	0.005	0.7915	0.2142
0.0123	16.0	24784	nan	0.005	0.7915	0.2142
0.009	17.0	26333	nan	0.005	0.7915	0.2142
0.0321	18.0	27882	nan	0.005	1.0	1.0
0.0	19.0	29431	nan	0.005	1.0	1.0

Framework versions

• Transformers 4.44.0
• Pytorch 2.2.0+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1