app.py

import gradio as gr
import re
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Modell und Tokenizer laden

# gpt2 outputs text!
#tokenizer = AutoTokenizer.from_pretrained("openai-community/gpt2")
#model = AutoModelForCausalLM.from_pretrained("openai-community/gpt2")

# distilgpt2 is only 80MB -> NOK, no emojis
tokenizer = AutoTokenizer.from_pretrained("distilgpt2")
model = AutoModelForCausalLM.from_pretrained("distilgpt2")

# tiny-gpt2 is only 20MB -> NOK, no emojis
#tokenizer = AutoTokenizer.from_pretrained("sshleifer/tiny-gpt2")
#model = AutoModelForCausalLM.from_pretrained("sshleifer/tiny-gpt2")

# TinyLlama
#tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")
#model = AutoModelForCausalLM.from_pretrained("TinyLlama/TinyLlama-1.1B-Chat-v1.0")

# OLD conversion method
def text_to_emoji_OLD(input_text):
    # Eingabetext bereinigen (optional)
    cleaned_text = re.sub(r"[.,!?;:]", "", input_text)

    # Prompt vorbereiten
    prompt = f'Convert the following sentence into an emoji-sequence which conveys a similar meaning and return only the emojis, no explanation:\n\n"{cleaned_text}"\n\n'

    # Tokenisierung und Generation
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(
        **inputs,
        max_new_tokens=30,
        do_sample=True,
        temperature=0.8,
        top_k=50
    )

    # Decodieren
    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Nur den Teil nach dem Prompt extrahieren
    emoji_part = generated_text[len(prompt):].strip()

    return emoji_part

# conversion method
def text_to_emoji(input_text):
    # Eingabetext bereinigen (optional)
    cleaned_text = re.sub(r"[.,!?;:]", "", input_text)

    # Pure pattern-based prompt
    prompt = (
        "Hi there → 👋🙂\n"
        "Good night → 🌙😴\n"
        "I love pizza → ❤️🍕\n"
        "It's raining → 🌧️☔\n"
        "Happy birthday → 🎉🎂🥳\n"
        "I am so tired → 😴💤\n"
        "Let’s go to the beach → 🏖️🌊😎\n"
        "I’m feeling lucky → 🍀🤞\n"
        "We’re getting married → 💍👰🤵\n"
        "Merry Christmas → 🎄🎁🎅\n"
        "Let’s party → 🎉🕺💃\n"
        f"{cleaned_text} →"
    )
    
    # Tokenisierung und Generation
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(
        **inputs,
        max_new_tokens=10,
        do_sample=True,
        temperature=0.9,
        top_k=50,
        pad_token_id=tokenizer.eos_token_id  # Prevents warning
    )

    # Decodieren
    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Nur den generierten Teil nach dem letzten "→"
    emoji_part = generated_text.split("→")[-1].strip().split("\n")[0]

    return emoji_part

# Gradio UI
iface = gr.Interface(
    fn=text_to_emoji,
    inputs=gr.Textbox(lines=2, placeholder="Enter a sentence..."),
    outputs="text",
    title="AI-Powered Emoji Translator",
    description="Enter a sentence, and the AI will transform it into an emoji-version 🥳"
)

iface.launch()