app.py

import gradio as gr
import spaces
from inference import SentimentClassifier
from huggingface_hub import snapshot_download
from huggingface_hub import hf_hub_download
import os

MODEL_REPO = "vuminhtue/qwen3_sentiment_tinystories"
FILENAME   = "Qwen3_200k_model_params.pt"

LOCAL_DIR = os.path.join(os.getcwd(), "models", "qwen3_sentiment_tinystories")

weights_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=FILENAME,
    local_dir="qwen3_sentiment_tinystories",  # any folder in runtime
    local_dir_use_symlinks=None  # not needed anymore; safe to omit
)


# Load classifier
classifier = SentimentClassifier(model_dir="qwen3_sentiment_tinystories",
                                 weights_path=weights_path)

def predict_sentiment(text):
    """Predict sentiment and return results"""
    result = classifier.predict(text)
    
    return {
        "Sentiment": result["sentiment"].upper(),
        "Confidence": f"{result['confidence']:.2%}",
        "Negative Probability": f"{result['probabilities']['negative']:.2%}",
        "Positive Probability": f"{result['probabilities']['positive']:.2%}"
    }

@spaces.GPU(duration=120)
def generate(prompt):
    return pipe(prompt).images
    
# Create interface
demo = gr.Interface(
    fn=predict_sentiment,
    inputs=gr.Textbox(
        label="Enter text to analyze",
        placeholder="Type your text here...",
        lines=3
    ),
    outputs=gr.JSON(label="Prediction Results"),
    title="🎭 Sentiment Analyzer",
    description="Classify text as positive or negative using Qwen3:0.6B embeddings + Logistic Regression",
    examples=[
        ["This movie was absolutely wonderful!"],
        ["Terrible experience, complete waste of time."],
        ["It's okay, nothing special."]
    ]
)

demo.launch()