README.md

---
pipeline_tag: zero-shot-classification
library_name: transformers
datasets:
- Xerv-AI/RainDrop-DTS
---

# RainDrop - Zero-Shot Text Classification Model

<img src='https://raw.githubusercontent.com/Electroiscoding/super-waddle/refs/heads/main/Screenshot_20250802-174037.png' alt='RainDrop Logo'>

RainDrop is an experimental zero-shot text classification model designed to explore transformer-based dual-encoder architectures with contrastive learning. Currently trained on a small, test-scale dataset, it provides a foundation for future improvements toward robust zero-shot classification.

At this stage, **RainDrop is not multilingual and shows modest accuracy metrics.** It is intended primarily for research, experimentation, and development purposes rather than production use.

**Features**

- Trained on a small, test-scale dataset  
- Contrastive learning with shared transformer encoder  
- Compatible with Hugging Face Transformers library  

**Benchmarking Snapshots**

<img src='https://raw.githubusercontent.com/Electroiscoding/super-waddle/refs/heads/main/download%20(1).png' alt='Benchmark 1'>  
<img src='https://raw.githubusercontent.com/Electroiscoding/super-waddle/refs/heads/main/download%20(2).png' alt='Benchmark 2'>  
<img src='https://raw.githubusercontent.com/Electroiscoding/super-waddle/refs/heads/main/download.png' alt='Benchmark 3'>  

**How to Use**

```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer
from huggingface_hub import hf_hub_download
import os

repo_id = 'Xerv-AI/RainDrop-Demo-1'
model_dir = './downloaded_raindrop_model'
os.makedirs(model_dir, exist_ok=True)

files_to_download = [
    'config.json',
    'model.safetensors',
    'tokenizer.json',
    'tokenizer_config.json',
    'vocab.txt',
    'special_tokens_map.json',
    'artifacts.pkl',
    'training_args.bin'
]

for file_name in files_to_download:
    try:
        hf_hub_download(repo_id=repo_id, filename=file_name, local_dir=model_dir)
    except Exception as e:
        print(f"Could not download {file_name}: {e}")

# Function to load a RainDrop model
def load_raindrop_model(model_path):
    """
    Loads a RainDrop model for sequence classification.
    """
    tokenizer = AutoTokenizer.from_pretrained(model_path)
    model = AutoModelForSequenceClassification.from_pretrained(model_path)

    class RainDropModel:
        def __init__(self, model, tokenizer):
            self.model = model
            self.tokenizer = tokenizer

        def predict_single(self, text):
            inputs = self.tokenizer(text, return_tensors="pt")
            outputs = self.model(**inputs)
            logits = outputs.logits
            predicted_class_id = logits.argmax().item()
            predicted_label = self.model.config.id2label[predicted_class_id]
            confidence = logits.softmax(dim=1)[0][predicted_class_id].item()
            return predicted_label, confidence

    return RainDropModel(model, tokenizer)


try:
    loaded_model_from_hub = load_raindrop_model(model_dir)

    test_text = "Physics is cool."
    predicted_label, confidence = loaded_model_from_hub.predict_single(test_text)

    print(f"Text: '{test_text}'")
    print(f"Predicted Label: {predicted_label}")
    print(f"Confidence: {confidence:.4f}")

except Exception as e:
    print(f"Error loading or evaluating the model from the hub: {e}")
```

About

RainDrop is developed by Xerv AI as an early-stage research project to push the boundaries of zero-shot classification. It is a stepping stone towards a more accurate, multilingual model in the future.