# 1. Import the required packages
import torch
import gradio as gr

from typing import Dict
from transformers import pipeline

# 2. Define our function to use with our model.
def food_not_food_classifier(text: str) -> Dict[str, float]:
  # 2. Setup food not food text classifier
  food_not_food_classifier_pipeline = pipeline(task="text-classification",
                                                model="mrdbourke/learn_hf_food_not_food_text_classifier-distilbert-base-uncased",
                                                batch_size=32,
                                                device="cuda" if torch.cuda.is_available() else "cpu",
                                                top_k=None) # top_k=None => return all possible labels

  # 3. Get the outputs from our pipeline
  outputs = food_not_food_classifier_pipeline(text)[0]

  # 4. Format output for Gradio
  output_dict = {}
  for item in outputs:
    output_dict[item["label"]] = item["score"]

  return output_dict

# 3. Create a Gradio interface
description = """
A text classifier to determine if a sentence is about food or not food.

Fine-tuned from [DistilBERT](https://huggingface.co/distilbert/distilbert-base-uncased) a [dataset of LLM generated food/not_food image captions](https://huggingface.co/datasets/mrdbourke/learn_hf_food_not_food_image_captions).

See [source code](https://github.com/mrdbourke/learn-huggingface/blob/main/notebooks/hugging_face_text_classification_tutorial.ipynb).
"""

demo = gr.Interface(
    fn=food_not_food_classifier,
    inputs="text",
    outputs=gr.Label(num_top_classes=2),
    title="🍗🚫🥑 Food or Not Food Text Classifier",
    description=description,
    examples=[["I whipped up a fresh batch of code, but it to seems to have a syntax error"],
              ["A plate of pancakes and strawberry icing"]]
)

# 4. Launch the interface
if __name__ == "__main__":
  demo.launch()