finetuning-sentiment-model-3000-samples

This model is a fine-tuned version of distilbert-base-uncased on a 3,000-example subset of the IMDB dataset for binary sentiment classification (positive / negative movie reviews).

Model description

DistilBERT is a smaller, faster, and cheaper version of BERT. This model adds a sequence classification head (2 labels) on top of distilbert-base-uncased and fine-tunes it for sentiment analysis.

• Task: Text Classification (Sentiment Analysis)
• Base model: distilbert-base-uncased
• Language: English
• Labels: LABEL_0 = Negative, LABEL_1 = Positive

Training and evaluation data

The model was trained on a shuffled subset of the IMDB dataset (seed=42):

• Training set: 3,000 examples
• Evaluation set: 300 examples

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: AdamW (betas=(0.9,0.999), epsilon=1e-08)
• lr_scheduler_type: linear
• num_epochs: 2

Training results

On the 300-example evaluation set, the model achieves:

Metric	Value
Loss	0.3321
Accuracy	0.8700
F1	0.8713

Framework versions

• Transformers 5.16.1
• PyTorch 2.6.0+cpu
• Datasets 5.0.1
• Tokenizers 0.23.2

How to use

from transformers import pipeline

sentiment_model = pipeline(
    model="ousmane93/finetuning-sentiment-model-3000-samples"
)
sentiment_model(["I love this movie", "This movie sucks!"])

Limitations

• Fine-tuned on a small subset of IMDB (3,000 training / 300 evaluation examples), so performance may be lower than a model trained on the full 25,000-example dataset.
• Trained only on English-language movie reviews; performance on other domains, topics, or languages is not guaranteed.
• No comparison against other sentiment models has been performed as part of this project.