Built as part of the SentimentOps MLOps project — an end-to-end pipeline covering data processing, model training, experiment tracking with MLflow, REST API with FastAPI, CLI tooling, and Docker deployment.

Training

Base model: distilbert-base-uncased
Dataset: IMDB (25,000 train / 25,000 test)
Epochs: 3
Batch size: 16
Learning rate: 2e-5
Hardware: NVIDIA Tesla T4 (Google Colab)
Training time: ~50 minutes

Performance

Metric	Value
Test Accuracy	97.61%
vs TF-IDF baseline	+7%

Usage

python

from transformers import pipeline

classifier = pipeline(
    "sentiment-analysis",
    model="YOUR_USERNAME/imdb-distilbert-sentimentops"
)

result = classifier("This movie was absolutely fantastic!")
print(result)
# [{'label': 'POSITIVE', 'score': 0.9976}]

Comparison with Baseline

Model	Accuracy	Inference
TF-IDF + LogReg	90.0%	~1ms
DistilBERT (this)	97.6%	~50ms

Project

Part of SentimentOps — a full MLOps pipeline:

GitHub: https://github.com/as4d-4k/sentimentops
Portfolio: https://portfolio-steel-nine-aur.vercel.app

asadullahrehmann/imdb-distilbert-sentimentops

IMDB Sentiment Analysis — DistilBERT

Model Description

Training

Performance

Usage

Comparison with Baseline

Project

asadullahrehmann/
imdb-distilbert-sentimentops