Update app.py

app.py

@@ -1,99 +1,38 @@
-from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer
+from transformers import Pix2StructForConditionalGeneration, Pix2StructProcessor
 import gradio as gr
-import torch
-from concurrent.futures import ThreadPoolExecutor
-from threading import Lock
+from PIL import Image
 
-# Global cache and lock for thread-safety
-CACHE_SIZE = 100
-prediction_cache = {}
-cache_lock = Lock()
+# Load the pre-trained Pix2Struct model and processor
+model_name = "google/pix2struct-mathqa-base"
+model = Pix2StructForConditionalGeneration.from_pretrained(model_name)
+processor = Pix2StructProcessor.from_pretrained(model_name)
 
-# Mapping for sentiment labels from cardiffnlp/twitter-roberta-base-sentiment
-SENTIMENT_LABEL_MAPPING = {
-    "LABEL_0": "negative",
-    "LABEL_1": "neutral",
-    "LABEL_2": "positive"
-}
-
-def load_model(model_name):
-    """
-    Loads the model with 8-bit quantization if a GPU is available;
-    otherwise, loads the full model.
-    """
-    if torch.cuda.is_available():
-        model = AutoModelForSequenceClassification.from_pretrained(
-            model_name, load_in_8bit=True, device_map="auto"
-        )
-        tokenizer = AutoTokenizer.from_pretrained(model_name)
-        device = 0  # GPU index
-    else:
-        model = AutoModelForSequenceClassification.from_pretrained(model_name)
-        tokenizer = AutoTokenizer.from_pretrained(model_name)
-        device = -1
-    return pipeline("text-classification", model=model, tokenizer=tokenizer, device=device)
-
-# Load both models concurrently at startup.
-with ThreadPoolExecutor() as executor:
-    sentiment_future = executor.submit(load_model, "cardiffnlp/twitter-roberta-base-sentiment")
-    emotion_future = executor.submit(load_model, "bhadresh-savani/bert-base-uncased-emotion")
-
-sentiment_pipeline = sentiment_future.result()
-emotion_pipeline = emotion_future.result()
-
-def analyze_text(text):
-    # Check cache first (thread-safe)
-    with cache_lock:
-        if text in prediction_cache:
-            return prediction_cache[text]
-    
-    try:
-        # Run both model inferences in parallel.
-        with ThreadPoolExecutor() as executor:
-            future_sentiment = executor.submit(sentiment_pipeline, text)
-            future_emotion = executor.submit(emotion_pipeline, text)
-            sentiment_result = future_sentiment.result()[0]
-            emotion_result = future_emotion.result()[0]
-        
-        # Remap the sentiment label to a human-readable format if available.
-        raw_sentiment_label = sentiment_result.get("label", "")
-        sentiment_label = SENTIMENT_LABEL_MAPPING.get(raw_sentiment_label, raw_sentiment_label)
-        
-        # Format the output with rounded scores.
-        result = {
-            "Sentiment": {sentiment_label: round(sentiment_result['score'], 4)},
-            "Emotion": {emotion_result['label']: round(emotion_result['score'], 4)}
-        }
-    except Exception as e:
-        result = {"error": str(e)}
+# Function to solve handwritten math problems
+def solve_math_problem(image):
+    # Preprocess the image
+    inputs = processor(images=image, text="Solve the math problem:", return_tensors="pt")
     
-    # Update the cache in a thread-safe manner.
-    with cache_lock:
-        if len(prediction_cache) >= CACHE_SIZE:
-            prediction_cache.pop(next(iter(prediction_cache)))
-        prediction_cache[text] = result
+    # Generate the solution
+    predictions = model.generate(**inputs, max_new_tokens=100)
     
-    return result
+    # Decode the output
+    solution = processor.decode(predictions[0], skip_special_tokens=True)
+    return solution
 
-# Define the Gradio interface.
+# Gradio interface
 demo = gr.Interface(
-    fn=analyze_text,
-    inputs=gr.Textbox(placeholder="Enter your text here...", label="Input Text"),
-    outputs=gr.JSON(label="Analysis Results"),
-    title="🚀 Fast Sentiment & Emotion Analysis",
-    description="Optimized application that remaps sentiment labels and uses parallel processing.",
+    fn=solve_math_problem,
+    inputs=gr.Image(type="pil", label="Upload Handwritten Math Problem"),
+    outputs=gr.Textbox(label="Solution"),
+    title="Handwritten Math Problem Solver",
+    description="Upload an image of a handwritten math problem, and the model will solve it.",
     examples=[
-        ["I'm thrilled to start this new adventure!"],
-        ["This situation is making me really frustrated."],
-        ["I feel so heartbroken and lost."]
+        ["example1.jpg"],  # Add example images
+        ["example2.jpg"]
     ],
-    theme="soft",
-    allow_flagging="never"
+    theme="soft"
 )
 
-# Warm up the models with a sample input.
-_ = analyze_text("Warming up models...")
-
+# Launch the app
 if __name__ == "__main__":
-    # Bind to all interfaces for Hugging Face Spaces.
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+    demo.launch()