Add Hugging Face Serverless API option (Option 3)

ai_router.py

@@ -1,6 +1,7 @@
 from models.blip_model import blip_answer
 from models.reasoning_model import reasoning_answer
 from models.gemini_vision import gemini_vision_answer
+from models.hf_api_model import hf_api_reasoning
 
 try:
     from deep_translator import GoogleTranslator
@@ -21,6 +22,8 @@ def route_model(model_choice, image, question, lang="en"):
         cap, ans, exp = reasoning_answer(image, question)
     elif model_choice == "gemini":
         cap, ans, exp = gemini_vision_answer(image, question, lang)
+    elif model_choice == "hf_api":
+        cap, ans, exp = hf_api_reasoning(image, question, lang)
     else:
         cap, ans, exp = "Unknown", "Invalid", "Invalid"
 

config.py

@@ -12,4 +12,5 @@ GEMINI_API_KEY = os.getenv("GEMINI_API_KEY", "AIzaSyB1fMfnnp2etuOVWiLrecdMp3_0Gb
 
 # ===== Security Limits =====
 MAX_IMAGE_SIZE_MB = 5
-RATE_LIMIT_PER_MINUTE = 30
+RATE_LIMIT_PER_MINUTE = 30
+HF_TOKEN = os.getenv("HF_TOKEN", "")

models/hf_api_model.py

@@ -0,0 +1,72 @@
+import os
+import io
+import base64
+import requests
+from config import HF_TOKEN
+
+def encode_image(image):
+    buffered = io.BytesIO()
+    # Convert image to RGB if not already to prevent JPEG save errors
+    if image.mode != "RGB":
+        image = image.convert("RGB")
+    image.save(buffered, format="JPEG")
+    return base64.b64encode(buffered.getvalue()).decode('utf-8')
+
+def hf_api_reasoning(image, question, lang="en"):
+    # Using Llama 3.2 11B Vision Instruct via Serverless
+    model_id = "meta-llama/Llama-3.2-11B-Vision-Instruct"
+    api_url = f"https://api-inference.huggingface.co/models/{model_id}/v1/chat/completions"
+    
+    token = HF_TOKEN or os.getenv("HF_TOKEN")
+    
+    if not token:
+        return "Setup Required", "Hugging Face Access Token is missing.", "Please ensure HF_TOKEN is set in your environment or config.py"
+        
+    headers = {
+        "Authorization": f"Bearer {token}",
+        "Content-Type": "application/json"
+    }
+    
+    base64_img = f"data:image/jpeg;base64,{encode_image(image)}"
+    
+    payload = {
+        "model": model_id,
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "image_url", "image_url": {"url": base64_img}},
+                    {"type": "text", "text": f"Please provide an accurate answer to the following question. First, provide a very short, direct final answer. Second, provide a detailed explanation.\\nQuestion: {question}"}
+                ]
+            }
+        ],
+        "max_tokens": 250
+    }
+    
+    try:
+        response = requests.post(api_url, headers=headers, json=payload, timeout=60)
+        
+        if response.status_code == 429:
+            return "Rate Limited", "Hugging Face Serverless API is currently busy.", "You have hit the free tier rate limit. Please try again in a few minutes."
+        elif response.status_code == 503:
+            return "Model Loading", "The AI model is currently booting up on Hugging Face servers.", "It takes about 20-30 seconds to wake up the 11-Billion parameter model. Try again in 20 seconds!"
+            
+        response.raise_for_status()
+        data = response.json()
+        
+        raw_text = data["choices"][0]["message"]["content"]
+        
+        return "Llama-3.2-11B-Vision", raw_text, "Processed via Hugging Face Serverless GPU API"
+        
+    except requests.exceptions.Timeout:
+        return "Timeout", "The Hugging Face server took too long to respond.", "This can happen if the model is waking up. Please try again."
+    except Exception as e:
+        error_msg = str(e)
+        try:
+            # Attempt to parse detailed error from huggingface
+            error_data = response.json()
+            if "error" in error_data:
+                error_msg = error_data["error"]
+        except:
+            pass
+        return "API Error", "Connection failed.", f"{error_msg}"

requirements.txt

@@ -6,3 +6,4 @@ torch==2.2.1
 transformers==4.38.2
 google-genai==0.3.0
 deep-translator==1.11.4
+requests>=2.28.0

static/app.js

@@ -354,7 +354,7 @@ async function fetchLogs() {
         appState.logs = data.logs || [];
         appState.stats.total = appState.logs.length;
         appState.stats.local = appState.logs.filter(l => ['local', 'blip', 'reasoning'].includes(l.model.toLowerCase())).length;
-        appState.stats.gemini = appState.logs.filter(l => ['gemini', 'external'].includes(l.model.toLowerCase())).length;
+        appState.stats.gemini = appState.logs.filter(l => ['gemini', 'external', 'hf_api'].includes(l.model.toLowerCase())).length;
         
         updateDashboardView();
     } catch(err) {
@@ -378,7 +378,7 @@ function renderLogs(filter = '') {
         tr.innerHTML = `
             <td>${log.timestamp}</td>
             <td><span class="user-badge" style="color: var(--text-secondary);"><i class="fa-solid fa-user"></i> ${log.user}</span></td>
-            <td><span style="color: ${log.model.includes('gemini') ? 'var(--accent)' : 'var(--success)'}">${log.model.toUpperCase()}</span></td>
+            <td><span style="color: ${['gemini', 'hf_api'].some(m => log.model.includes(m)) ? 'var(--accent)' : 'var(--success)'}">${log.model.toUpperCase().replace('_', ' ')}</span></td>
             <td>${log.question.length > 50 ? log.question.substring(0, 50) + '...' : log.question}</td>
         `;
         els.logsBody.appendChild(tr);

templates/index.html

@@ -97,6 +97,7 @@
                                     <select id="model-selector">
                                         <option value="local">BLIP + FLAN (Local)</option>
                                         <option value="gemini" id="gemini-option">Gemini 3.1 Flash Vision (High Quota)</option>
+                                        <option value="hf_api" id="hf_api-option">HF Llama 3.2 Vision API (Flawless)</option>
                                     </select>
                                 </div>
                                 <div class="control-group" style="flex: 1;">