CS224S_Quechua_Project

This model is a fine-tuned version of cportoca/CS224S_Quechua_Project on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0264
• Wer: 0.6160

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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4218	0.625	45	1.2929	0.8437
0.4233	1.25	90	1.3785	0.8580
0.4098	1.875	135	1.2656	0.8277
0.4212	2.5	180	1.1368	0.7781
0.3174	3.125	225	1.1210	0.8134
0.2819	3.75	270	1.0151	0.7221
0.2226	4.375	315	1.0450	0.7723
0.2152	5.0	360	1.0446	0.7100
0.2023	5.625	405	1.0544	0.7339
0.1547	6.25	450	1.0352	0.6932
0.1358	6.875	495	1.0490	0.6562
0.1229	7.5	540	1.0429	0.6500
0.079	8.125	585	0.9882	0.6532
0.0896	8.75	630	1.0109	0.6322
0.052	9.375	675	1.0006	0.6275
0.0515	10.0	720	1.0264	0.6160

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1