README.md

---
title: Meme vs Real Event Classifier
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
---

# Meme vs Real Event Tweet Classifier

Streamlit demo for a fine-tuned `bert-base-uncased` model that classifies a
tweet as a **meme / low-signal post** or a **real-world event**.

The model weights live in a separate Hugging Face model repo and are loaded
at startup via `transformers.AutoModelForSequenceClassification.from_pretrained`.

## Configure the model repo

The app reads the model id from the `MODEL_ID` environment variable, defaulting
to `Aryan047/Dynamic-event-detector`. To override in the Space UI go to
**Settings -> Variables and secrets** and set `MODEL_ID` to any other model repo.

## Local development

```bash
pip install -r requirements.txt
streamlit run app.py
```

## Files

- `app.py` - Streamlit application (single-tweet tab, batch-CSV tab)
- `requirements.txt` - runtime dependencies
- `upload_model.py` - one-shot helper to push `artifacts_meme_vs_event/bert_classifier/`
  to a new Hugging Face model repo. Not used by the Space itself.