chennab28/movie-review-sentiment-classifier

🍿 Movie Review Sentiment Classifier

A tiny BERT model (google/bert_uncased_L-4_H-256_A-4, ~10M params) fine-tuned for 3-way movie review sentiment classification:

positive / negative / neutral

Trained on the companion dataset chennab28/movie-review-sentiment in a couple of minutes on CPU.

Quick inference

from transformers import pipeline

pipe = pipeline("text-classification", model="chennab28/movie-review-sentiment-classifier")

print(pipe("A breathtaking masterpiece with stunning cinematography."))
# [{'label': 'positive', 'score': 0.99}]

Metrics

Task	Acc (test)
Sentiment (3 classes)	97.6%

Why so small?

This is deliberately a minimal end-to-end example of the Hugging Face workflow — Dataset → Model → Space. Despite having only ~10M parameters, the model reaches 97.6% accuracy on the held-out test split.

Training details

Config	Value
base model	google/bert_uncased_L-4_H-256_A-4
epochs	30
batch size	16
lr	5e-5
max seq len	128
data	168 train / 42 test rows

Reproduce with train/train.py --epochs 30 in this repo.

Try it live

👉 Gradio demo Space