app.py

import torch
import gradio as gr
from transformers import (
    AutoModelForSequenceClassification,
    AutoTokenizer,
)
from typing import Dict
import os
import pandas as pd
from huggingface_hub import login
login(token=os.getenv("HUGGINGFACE_TOKEN"))

FOUNDATIONS = ["authority", "care", "fairness", "loyalty", "sanctity"]
tokenizer = AutoTokenizer.from_pretrained(
    "joshnguyen/mformer-authority",
    use_auth_token=True
)
DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
MODELS = {}
for foundation in FOUNDATIONS:
    model = AutoModelForSequenceClassification.from_pretrained(
        pretrained_model_name_or_path=f"joshnguyen/mformer-{foundation}",
        use_auth_token=True
    )
    model.eval()
    MODELS[foundation] = model.to(DEVICE)


def classify_text(text: str) -> Dict[str, float]:
    # Encode the prompt
    inputs = tokenizer([text],
                       padding=True,
                       truncation=True,
                       return_tensors='pt').to(DEVICE)
    scores = []
    for foundation in FOUNDATIONS:
        model = MODELS[foundation]
        outputs = model(**inputs)
        outputs = torch.softmax(outputs.logits, dim=1)
        outputs = outputs[:, 1]
        score = outputs.detach().cpu().numpy()[0]
        scores.append([foundation.capitalize(), score])
    scores = pd.DataFrame(scores, columns=["foundation", "score"])
    return scores


demo = gr.Interface(
    fn=classify_text,
    inputs=[
        # Prompt
        gr.Textbox(
            label="Input text",
            container=False,
            show_label=True,
            placeholder="Enter some text...",
            lines=12,
            scale=10,
        ),
    ],
    outputs=[
        gr.BarPlot(
            x="foundation",
            y="score",
            title="Moral foundations scores",
            x_title=" ",
            y_title=" ",
            y_lim=[0, 1],
            vertical=False,
            height=200,
            width=500,
        )
    ],
)

demo.queue(max_size=20).launch()