Update app.py

app.py

@@ -1,6 +1,3 @@
-from fastapi import FastAPI, HTTPException
-from fastapi.middleware.cors import CORSMiddleware
-from pydantic import BaseModel
 import gradio as gr
 from huggingface_hub import InferenceClient
 from PIL import Image, ImageEnhance, ImageFilter
@@ -12,7 +9,6 @@ import os
 import base64
 from io import BytesIO
 import math
-import uvicorn
 
 # Global storage
 pdf_texts = {}
@@ -20,30 +16,6 @@ reader = None
 blip_processor = None
 blip_model = None
 
-# === FASTAPI APP ===
-app = FastAPI(title="AI Assistant API")
-
-# Add CORS
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-# Request model
-class AnalyzeRequest(BaseModel):
-    message: str
-    image_base64: str = None
-    model: str = "Qwen/Qwen2.5-7B-Instruct"
-
-# Response model
-class AnalyzeResponse(BaseModel):
-    analysis: str
-
-# === HELPER FUNCTIONS ===
-
 def load_pdfs():
     global pdf_texts
     pdf_texts.clear()
@@ -98,6 +70,14 @@ def decode_base64_image(image_data):
             pass
     return image_data
 
+def web_search(query):
+    try:
+        from ddgs import DDGS
+        results = DDGS().text(query, max_results=2)
+        return "\n".join([f"{r['title']}: {r['body'][:100]}" for r in results])
+    except:
+        return None
+
 def analyze_image(image):
     initialize_vision_models()
     try:
@@ -116,98 +96,125 @@ def analyze_image(image):
     except Exception as e:
         return "", str(e)
 
-# === FASTAPI ENDPOINTS ===
-
-@app.get("/")
-def read_root():
-    return {"message": "AI Assistant API is running", "endpoints": ["/api/analyze", "/gradio"]}
+def extract_math_calcs(text):
+    calcs = []
+    for match in re.finditer(r'C\s*\(\s*(\d+)\s*,\s*(\d+)\s*\)', text):
+        n, k = int(match.group(1)), int(match.group(2))
+        result = math.comb(n, k)
+        calcs.append(f"C({n},{k})={result:,}")
+    return calcs
 
-@app.get("/health")
-def health_check():
-    return {"status": "healthy"}
+def get_pdf_context(query):
+    if not pdf_texts:
+        return None
+    keywords = set(re.findall(r'\b\w{4,}\b', query.lower()))
+    chunks = []
+    for path, text in pdf_texts.items():
+        for sent in text.split('.')[:40]:
+            score = sum(1 for kw in keywords if kw in sent.lower())
+            if score > 0:
+                chunks.append((score, sent[:150]))
+    chunks.sort(reverse=True)
+    if chunks and chunks[0][0] >= 2:
+        return chunks[0][1]
+    return None
 
-@app.post("/api/analyze", response_model=AnalyzeResponse)
-async def analyze_endpoint(request: AnalyzeRequest):
+# MAIN API FUNCTION - Simple interface for external calls
+def api_analyze(message: str, image_base64: str = None, model: str = "Qwen/Qwen2.5-7B-Instruct"):
     """
-    REST API endpoint for analyzing text/images
+    Simple API function for external calls
     """
-    try:
-        token = os.getenv('HF_TOKEN')
-        message = request.message
-        image_base64 = request.image_base64
-        model = request.model
-        
-        # Decode image
-        img = None
-        if image_base64:
-            img = decode_base64_image(image_base64)
-        
-        # Detect MCQ
-        has_options = bool(re.search(r'[A-D][\.\)]\s', message))
-        
-        # Get context from image
-        context = ""
-        if img:
-            try:
-                ocr_text, _ = analyze_image(img)
-                if ocr_text:
-                    context = f"\n\nExtracted text:\n{ocr_text[:400]}"
-            except Exception as e:
-                context = f"\n\n(OCR error: {str(e)})"
-        
-        # System message
-        if has_options:
-            sys_msg = "Exam assistant. Format: Answer: [letter]. Reason: [one sentence]."
-            temp = 0.2
-            tokens = 100
-        else:
-            sys_msg = "You are a helpful AI assistant."
-            temp = 0.6
-            tokens = 400
-        
+    token = os.getenv('HF_TOKEN')
+    
+    # Decode image
+    img = None
+    if image_base64:
+        img = decode_base64_image(image_base64)
+    
+    # Detect MCQ
+    has_options = bool(re.search(r'[A-D][\.\)]\s', message))
+    
+    # Get context from image
+    context = ""
+    if img:
         try:
-            client = InferenceClient(token=token, model=model)
-        except:
-            try:
-                client = InferenceClient(token=token, model="Qwen/Qwen2.5-7B-Instruct")
-            except Exception as e:
-                raise HTTPException(status_code=500, detail=f"Model connection failed: {str(e)}")
-        
-        messages = [
-            {"role": "system", "content": sys_msg},
-            {"role": "user", "content": message + context}
-        ]
-        
-        # Non-streaming response
-        response = ""
+            ocr_text, _ = analyze_image(img)
+            if ocr_text:
+                context = f"\n\nExtracted text from image:\n{ocr_text[:400]}"
+        except Exception as e:
+            context = f"\n\n(Image processing error: {str(e)})"
+    
+    # System message
+    if has_options:
+        sys_msg = "You are an exam assistant. For MCQ, give: Answer: [letter]. Reason: [one sentence only]."
+        temp = 0.2
+        tokens = 100
+    else:
+        sys_msg = "You are a helpful AI assistant."
+        temp = 0.6
+        tokens = 400
+    
+    try:
+        client = InferenceClient(token=token, model=model)
+    except:
         try:
-            for msg in client.chat_completion(messages, max_tokens=tokens, stream=True, temperature=temp, top_p=0.9):
-                if msg.choices and msg.choices[0].delta.content:
-                    response += msg.choices[0].delta.content
-                    if has_options and len(response) > 250:
-                        break
+            client = InferenceClient(token=token, model="Qwen/Qwen2.5-7B-Instruct")
         except Exception as e:
-            raise HTTPException(status_code=500, detail=f"Inference failed: {str(e)}")
-        
-        return AnalyzeResponse(analysis=response.strip())
-        
-    except HTTPException:
-        raise
+            return f"Error connecting to model: {str(e)}"
+    
+    messages = [
+        {"role": "system", "content": sys_msg},
+        {"role": "user", "content": message + context}
+    ]
+    
+    # Non-streaming response
+    response = ""
+    try:
+        for msg in client.chat_completion(
+            messages, 
+            max_tokens=tokens, 
+            stream=True, 
+            temperature=temp, 
+            top_p=0.9
+        ):
+            if msg.choices and msg.choices[0].delta.content:
+                response += msg.choices[0].delta.content
+                
+                # Stop early for MCQ
+                if has_options and len(response) > 200:
+                    break
     except Exception as e:
-        raise HTTPException(status_code=500, detail=str(e))
-
-# === GRADIO UI ===
+        return f"Error during inference: {str(e)}"
+    
+    return response.strip()
 
-def respond(message, history, system_message, max_tokens, temperature, top_p, model_selection, image, hf_token):
-    """Gradio chat function"""
+# Chat function for UI
+def respond(
+    message,
+    history: list[dict[str, str]],
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+    model_selection,
+    image,
+    hf_token,
+):
+    """UI chat function with streaming"""
     if image is not None:
         image = decode_base64_image(image)
     
     token = os.getenv('HF_TOKEN') or (hf_token.strip() if hf_token else None)
     has_options = bool(re.search(r'[A-D][\.\)]\s', message))
+    is_math_calc = any(w in message.lower() for w in ['calculate', 'factorial', 'combination'])
+    
+    if is_math_calc and not has_options:
+        selected_model = "Qwen/Qwen2.5-Math-7B-Instruct"
+    else:
+        selected_model = model_selection
     
     try:
-        client = InferenceClient(token=token, model=model_selection)
+        client = InferenceClient(token=token, model=selected_model)
     except:
         try:
             client = InferenceClient(token=token, model="Qwen/Qwen2.5-7B-Instruct")
@@ -225,7 +232,7 @@ def respond(message, history, system_message, max_tokens, temperature, top_p, mo
             pass
     
     if has_options:
-        system_message = "Exam assistant. Answer: [letter]. Reason: [one sentence]."
+        system_message = "Exam assistant. MCQ format: Answer: [letter]. Reason: [one sentence]."
         temperature = 0.2
         max_tokens = 100
     
@@ -246,7 +253,8 @@ def respond(message, history, system_message, max_tokens, temperature, top_p, mo
 # Load PDFs
 pdf_status = load_pdfs()
 
-# Create Gradio interface
+# Create TWO separate interfaces
+# 1. Chat UI for users
 chat_interface = gr.ChatInterface(
     respond,
     type="messages",
@@ -256,7 +264,7 @@ chat_interface = gr.ChatInterface(
         gr.Slider(0.1, 1.2, 0.6, step=0.1, label="Temperature"),
         gr.Slider(0.1, 1.0, 0.9, step=0.05, label="Top-p"),
         gr.Dropdown(
-            choices=["Qwen/Qwen2.5-7B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "HuggingFaceH4/zephyr-7b-beta"],
+            choices=["Qwen/Qwen2.5-7B-Instruct", "meta-llama/Llama-3.2-3B-Instruct", "HuggingFaceH4/zephyr-7b-beta","openai/gpt-oss-20b","Qwen/Qwen2.5-Math-7B-Instruct"],
             value="Qwen/Qwen2.5-7B-Instruct",
             label="Model",
         ),
@@ -267,10 +275,26 @@ chat_interface = gr.ChatInterface(
     description=f"MCQ (short) • Math (steps) • General (detailed)\n\n{pdf_status}",
 )
 
-# Mount Gradio to FastAPI
-app = gr.mount_gradio_app(app, chat_interface, path="/gradio")
+# 2. Simple API interface
+api_interface = gr.Interface(
+    fn=api_analyze,
+    inputs=[
+        gr.Textbox(label="Message", placeholder="Enter your question"),
+        gr.Textbox(label="Image (base64)", placeholder="Optional base64 image"),
+        gr.Textbox(label="Model", value="Qwen/Qwen2.5-7B-Instruct"),
+    ],
+    outputs=gr.Textbox(label="Response"),
+    title="API Endpoint",
+    description="Direct API access",
+    api_name="analyze"  # Creates /call/analyze endpoint
+)
 
-# === RUN SERVER ===
+# Combine both in tabs
+demo = gr.TabbedInterface(
+    [chat_interface, api_interface],
+    ["Chat", "API"],
+    title="🤖 AI Assistant"
+)
 
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    demo.launch()