W2V2_Bert_BIG_C_Bemba_100hr_v1

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4908
• Wer: 0.3467
• Cer: 0.0931

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-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.025
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.5063	1.0	12851	0.6450	0.4944	0.1202
0.6656	2.0	25702	0.5112	0.4189	0.1037
0.6012	3.0	38553	0.4759	0.3943	0.0999
0.5586	4.0	51404	0.4518	0.3684	0.0930
0.5269	5.0	64255	0.4608	0.3510	0.0912
0.5006	6.0	77106	0.4594	0.3449	0.0885
0.4764	7.0	89957	0.4323	0.3358	0.0872
0.452	8.0	102808	0.4257	0.3465	0.0903
0.4295	9.0	115659	0.4303	0.3328	0.0858
0.4064	10.0	128510	0.4404	0.3272	0.0854
0.3823	11.0	141361	0.4655	0.3291	0.0855
0.3591	12.0	154212	0.4748	0.3312	0.0859
0.3352	13.0	167063	0.4645	0.3405	0.0919
0.3127	14.0	179914	0.5077	0.3317	0.0860
0.2897	15.0	192765	0.4963	0.3370	0.0879
0.2686	16.0	205616	0.5166	0.3373	0.0882
0.2482	17.0	218467	0.5365	0.3382	0.0883
0.2289	18.0	231318	0.5852	0.3401	0.0883
0.2101	19.0	244169	0.6336	0.3415	0.0889
0.193	20.0	257020	0.6719	0.3402	0.0884

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.0
• Tokenizers 0.19.1