app.py

import gradio as gr
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
import torch.nn.functional as F
from huggingface_hub import login
import os

# Authenticate with Hugging Face using token from environment variable
login(os.environ.get("HUGGINGFACE_TOKEN"))

# Load MentalBERT model & tokenizer
MODEL_NAME = "mental/mental-bert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSequenceClassification.from_pretrained(
    MODEL_NAME,
    num_labels=2,
    problem_type="single_label_classification"
)

LABELS = {
    "neutral": {"index": 0, "description": "Emotionally balanced or calm"},
    "emotional": {"index": 1, "description": "Showing emotional content"}
}

def analyze_text(text):
    # Tokenize input
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    
    # Get model predictions
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probs = F.softmax(logits, dim=-1)[0]

    # Get emotion scores
    emotions = {
        label: float(probs[info["index"]])
        for label, info in LABELS.items()
    }
    
    return emotions

# Create Gradio interface
iface = gr.Interface(
    fn=analyze_text,
    inputs=gr.Textbox(label="Enter text to analyze", lines=3),
    outputs=gr.Json(label="Emotion Analysis"),
    title="MentalBERT Emotion Analysis",
    description="Analyze the emotional content of text using MentalBERT (specialized for mental health content)",
    examples=[
        ["I feel really anxious about my upcoming presentation"],
        ["I've been feeling quite depressed lately"],
        ["I'm managing my stress levels well today"],
        ["Just had a great therapy session!"]
    ]
)

# Launch the interface
iface.launch()