Zk1-QA-VN-test2 (ZycckZ/Simple_VieQA)

This model is a fine-tuned version of ZycckZ/Zk1-QA-VN-test and taidng/UIT-ViQuAD2 dataset.

It achieves the following results on the evaluation set:

• Loss: 1.8800
• Exact Match (EM): 70.07
• F1 Score: 82.34

Model description

This model now much better than the model before (ZycckZ/Zk1-QA-VN-test).

Intended uses & limitations

• Create a simple chatbot QA 🤗
• Just for Vietnamese QA system

Training and evaluation data

Training:

• taidng/UIT-ViQuAD2.0 - "train"
• taidng/UIT-ViQuAD2.0 - "validation"
• taidng/UIT-ViQuAD2.0 - "test"

Evaluation:

• taidng/UIT-ViQuAD2.0 - "validation"

Training procedure

Based on Question Answering HuggingFace 🤗

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss
1.171	1.0	2226	1.0648
0.8499	2.0	4452	1.1054
0.5862	3.0	6678	1.2583
0.4082	4.0	8904	1.5642
0.2835	5.0	11130	1.8800

Framework versions

• Transformers 4.52.2
• Pytorch 2.6.0+cu124
• Datasets 2.14.4
• Tokenizers 0.21.1