vishalp23/distilbert-subject-classifier

Subject Classifier built on Distilbert

Table of Contents

• Model Details
• How to Get Started With the Model
• Uses
• Risks, Limitations and Biases
• Training
• Evaluation
• Environmental Impact

Model Details

Model Description: This is the uncased DistilBERT model fine-tuned on a custom dataset that is built on the IITJEE NEET AIIMS Students Questions Data for the subject classification task.

• Developed by: The Typeform team.
• Model Type: Text Classification
• Language(s): English
• License: GNU GENERAL PUBLIC LICENSE
• Parent Model: See the distilbert base uncased model for more information about the Distilled-BERT base model.

Uses

This model can be used for text classification tasks.

Risks, Limitations and Biases

CONTENT WARNING: Readers should be aware this section contains content that is disturbing, offensive, and can propagate historical and current stereotypes.

Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)).

Training

Training is done on a NVIDIA RTX 3070 AMD Ryzen 7 5800 with the following hyperparameters:

$ training.ipynb \
    --model_name_or_path distilbert-base-uncased \
    --do_train \
    --do_eval \
    --max_seq_length 512 \
    --per_device_train_batch_size 4 \
    --learning_rate 1e-05 \
    --num_train_epochs 5 \

Evaluation

Evaluation Results

When fine-tuned on downstream tasks, this model achieves the following results:

Epochs: 5 | Train Loss: 0.001 | Train Accuracy: 0.989 | Val Loss: 0.006 | Val Accuracy: 0.950 CPU times: user 18h 19min 13s, sys: 1min 34s, total: 18h 20min 47s Wall time: 18h 20min 7s

• **Epoch = ** 5.0
• Evaluation Accuracy = 0.950
• Evaluation Loss = 0.006
• Training Accuracy = 0.989
• Training Loss = 0.001

Testing Results

	precision	recall	f1-score	support
biology	0.98	0.99	0.99	15988
chemistry	1.00	0.99	0.99	20678
computer	1.00	0.99	0.99	8754
maths	1.00	1.00	1.00	26661
physics	0.99	0.98	0.99	10306
social sciences	0.99	1.00	0.99	25695
accuracy	0.99	108082
macro avg	0.99	0.99	0.99	108082
weighted avg	0.99	0.99	0.99	108082

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). We present the hardware type based on the associated paper.

Hardware Type: 1 NVIDIA RTX 3070

Hours used: 18h 19min 13s

Carbon Emitted: (Power consumption x Time x Carbon produced based on location of power grid): Unknown