app.py

import os
from transformers import TFBertForSequenceClassification, BertTokenizerFast

def load_model(model_name):
    try:
        # Load TensorFlow model from Hugging Face
        model = TFBertForSequenceClassification.from_pretrained(model_name, use_auth_token=os.getenv('API_KEY'))
    except OSError:
        # Fallback to PyTorch model if TensorFlow fails
        model = TFBertForSequenceClassification.from_pretrained(model_name, use_auth_token=os.getenv('API_KEY'), from_pt=True)
    return model

def load_tokenizer(model_name):
    tokenizer = BertTokenizerFast.from_pretrained(model_name, use_auth_token=os.getenv('API_KEY'))
    return tokenizer

def predict(text, model, tokenizer):
    inputs = tokenizer(text, return_tensors="tf")
    outputs = model(**inputs)
    return outputs

def main():
    model_name = os.getenv('MODEL_PATH')
    if model_name is None:
        raise ValueError("MODEL_PATH environment variable not set or is None")

    model = load_model(model_name)
    tokenizer = load_tokenizer(model_name)

    # Example prediction
    text = "Sample input text"
    result = predict(text, model, tokenizer)
    print(result)

if __name__ == "__main__":
    main()