# Crypto Project Classifier  

This is a Hugging Face model built on **FacebookAI/roberta-large**, fine-tuned to classify Twitter accounts as **crypto projects** or **non-crypto entities**. The model takes as input a sentence containing the author's name and bio, and outputs a probability score and a classification (1 for a crypto project, 0 otherwise).  

## Model Details  

- **Language**: English  
- **Base Model**: FacebookAI/roberta-large  
- **Task**: Sequence Classification  

### Example Input  

```plaintext  
hi i am {author_name}, i do this {twitter_bio}  
```  

## How to Use  

Below is a sample Python script to use the model for classification:  

```python  
import torch  
import pandas as pd  
from transformers import RobertaTokenizer, AutoModelForSequenceClassification  

model_name = "yoursdevkalki/crypto_project_classifier"  
tokenizer = RobertaTokenizer.from_pretrained(model_name)  
model = AutoModelForSequenceClassification.from_pretrained(model_name)  

projects_df = pd.read_csv("projects.csv")  
test_df = projects_df.head(500)  

def process_row(description):  
    inputs = tokenizer(description, return_tensors="pt", padding=True, truncation=True)  

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

    logits = outputs.logits  
    probability = torch.sigmoid(logits).numpy()[0][0]  # Convert to probability  

    # Compute prediction  
    prediction = 1 if probability >= 0.6 else 0  

    return probability * 100, prediction  

test_df["prob"], test_df["prediction"] = zip(*test_df["twitter_bio"].apply(process_row))  

print(test_df)  
```  

### Input Format  

The input text should be structured as:  
```plaintext  
hi i am {author_name}, i do this {twitter_bio}  
```  

### Outputs  

- **prob**: The model's confidence in percentage (0–100%).  
- **prediction**: Classification result (`1` for crypto project, `0` for non-project).  

## Dataset  

The model was trained on a dataset of 40k samples:  
- **20k Crypto Projects** (labeled as `1`)  
- **20k Non-Crypto Entities** (labeled as `0`)  

### Metrics Achieved  

- **F1 Score**: >90%  
- **Accuracy**: >90%  

## Donations  

If you find this model useful, consider supporting its development:  

- **Solana Address**: `2oiBTZ3QvTbsns4babAW54PHcKzacYG3MXUcpAMp7LKV`  
- **Ethereum Address**: `0x56a28F1Bd2CD4E2AAA386aeA1c30a24A2f854Ec4`  

## Reach Out  

- **Twitter**: [@yourdevkalki](https://x.com/yourdevkalki)  
- **Email**: `yourdevkalki@gmail.com`