qa_en_ms_model_v4

This model is a fine-tuned version of timpal0l/mdeberta-v3-base-squad2 on self collected dataset. It achieves the following results on the evaluation set:

• Loss: 0.7732

Intended uses & limitations

This should only be used for question answering task. The model needs context to be able to answer user's question.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• 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: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.8094	1.0	3139	0.7540
0.6089	2.0	6278	0.6963
0.5022	3.0	9417	0.7732

Framework versions

• Transformers 4.52.4
• Pytorch 2.5.1+cu121
• Datasets 3.6.0
• Tokenizers 0.21.1