distilbert-science-exam-sm

This model is a fine-tuned version of distilbert-base-cased on the LLM Science Exam dataset from Kaggle for MCQA. It achieves the following results on the evaluation set:

• Loss: 1.2390
• Accuracy: 0.7

Intended uses & limitations

Multiple Choice Question & Answer on Medical Topics

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	1.0	6	1.0428	0.5
No log	2.0	12	1.3253	0.5
No log	3.0	18	1.2390	0.7

Framework versions

• Transformers 4.31.0
• Pytorch 2.0.1
• Datasets 2.14.0
• Tokenizers 0.13.3