CS224S_Quechua_Project_Bilingual

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2367
• Wer: 0.2585

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: linear
• lr_scheduler_warmup_steps: 70
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.2273	0.3628	600	0.6478	0.6345
0.5989	0.7255	1200	0.4562	0.4218
0.4847	1.0883	1800	0.3781	0.3914
0.4599	1.4510	2400	0.3657	0.3400
0.3462	1.8138	3000	0.3296	0.3185
0.3738	2.1765	3600	0.2808	0.2975
0.2969	2.5393	4200	0.2856	0.2877
0.3985	2.9021	4800	0.2728	0.2889
0.2507	3.2648	5400	0.2676	0.2732
0.284	3.6276	6000	0.2539	0.2553
0.317	3.9903	6600	0.2359	0.2496
0.1526	4.3531	7200	0.2444	0.2609
0.1813	4.7158	7800	0.2367	0.2585

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1