Fine-tuned DistilBERT with LoRA for Sentiment Classification

This model is a lightweight DistilBERT fine-tuned with LoRA adapters on the IMDB movie-review dataset.
It predicts whether a given review is Positive or Negative and is used inside the Self-Healing Classification Pipeline project.


Model Details

  • Base model: distilbert-base-uncased
  • Fine-tuning method: LoRA (Low-Rank Adaptation)
  • Dataset: IMDB (50 k movie reviews)
  • Labels: 0 = Negative, 1 = Positive
  • Framework: 🤗Transformers / PEFT
  • Trained in: Google Colab GPU

Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("ranjana1811/fine-tuned-distilbert-lora")
model = AutoModelForSequenceClassification.from_pretrained("ranjana1811/fine-tuned-distilbert-lora")

text = "I loved every moment, the acting was superb!"
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
pred = outputs.logits.softmax(dim=-1).argmax().item()
print(["Negative","Positive"][pred])

Intended Use

  • Sentiment analysis for short texts or movie reviews.
  • Educational demo for explainable AI and LangGraph workflows.
  • Designed to be lightweight and run easily on CPU.

Not intended for: hate-speech detection, multi-label, or domain-specific tasks beyond simple sentiment.


Training Summary

Parameter Value
Epochs 3
Batch Size 16
Learning Rate 2e-5
Max Seq Len 256
LoRA Rank (r) 8
LoRA Alpha 16

Model trained with Trainer API on a small IMDB subset for faster Colab runs.


Evaluation (Validation subset)

Metric Score
Accuracy ≈ 0.90
F1 Score ≈ 0.90

Limitations & Risks

  • Can be over-confident on very short texts.
  • Works best on English movie reviews; not tested on other domains.
  • Does not filter biased or offensive content.

Environmental Note

Trained for ≈ 20 minutes on a T4 GPU in Google Colab – low carbon footprint.


Project Links


Author

Ranjana ( @ranjana1811 ) For internship submission – ATG Machine Learning Assignment 2025.


---
Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Dataset used to train ranjana1811/fine-tuned-distilbert-lora