A fine-tuned DistilBERT model for binary sentiment analysis on movie reviews. Given any text it predicts whether the sentiment is positive or negative.

Model Details

Attribute	Value
Base model	`distilbert-base-uncased`
Fine-tuned on	IMDb Movie Reviews
Task	Text Classification (Sentiment Analysis)
Labels	`positive` (1) / `negative` (0)
Max sequence length	512 tokens
Framework	PyTorch + 🤗 Transformers
License	Apache 2.0

Performance

Evaluated on the IMDb test split (25 000 samples):

Metric	Score
Accuracy	~93.2%
F1 (binary)	~93.1%

Quick Start

python

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="ohanvi/ohanvi-sentiment-analysis",
)

result = classifier("This movie was absolutely fantastic!")
# → [{'label': 'positive', 'score': 0.9978}]

result = classifier("Terrible film, complete waste of time.")
# → [{'label': 'negative', 'score': 0.9965}]

Training Details

Hyperparameters

Parameter	Value
Epochs	3
Batch size (train)	16
Learning rate	2e-5
Weight decay	0.01
Warmup ratio	10%
Optimiser	AdamW
LR scheduler	Linear with warmup

Training Data

The model was fine-tuned on the full IMDb dataset:

Train: 25 000 reviews (12 500 positive, 12 500 negative)
Test: 25 000 reviews (12 500 positive, 12 500 negative)

Training Environment

Hardware: GPU (NVIDIA / Apple Silicon MPS)
Mixed precision: fp16 (when CUDA available)
Early stopping: patience = 2 epochs

How to Use (Advanced)

python

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "ohanvi/ohanvi-sentiment-analysis"
tokenizer  = AutoTokenizer.from_pretrained(model_name)
model      = AutoModelForSequenceClassification.from_pretrained(model_name)
model.eval()

text   = "An outstanding film with incredible performances."
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

with torch.no_grad():
    logits = model(**inputs).logits

probs      = torch.softmax(logits, dim=-1)
label_id   = probs.argmax().item()
label      = model.config.id2label[label_id]
confidence = probs[0][label_id].item()

print(f"Label: {label}  ({confidence:.1%})")

Limitations

Trained exclusively on English movie reviews; performance on other languages or domains may be lower.
Very short texts (< 5 words) may produce less reliable results.
The model inherits any biases present in the IMDb dataset.

Citation

If you use this model, please cite:

bibtex

@misc{ohanvi-sentiment-2026,
  title   = {Ohanvi Sentiment Analysis},
  author  = {Gourav Bansal},
  year    = {2026},
  url     = {https://huggingface.co/ohanvi/ohanvi-sentiment-analysis},
}

Acknowledgements

Built with 🤗 Transformers, 🤗 Datasets, and Gradio.