Update app.py

app.py

@@ -12,6 +12,10 @@ try:
     SPACES_ENABLED = True
 except ImportError:
     SPACES_ENABLED = False
+    class spaces:
+        @staticmethod
+        def GPU(f):
+            return f
 
 # All dependencies are installed via requirements.txt / apt.txt
 from pypdf import PdfReader
@@ -35,22 +39,25 @@ _MODELS = [
 def call_llama(prompt: str, max_tokens: int = 2000) -> str:
     token = os.environ.get("HF_TOKEN", "")
     headers = {"Authorization": f"Bearer {token}"} if token else {}
+    
+    # Add default header to prevent gating issues
+    headers["User-Agent"] = "Phoenix-Framework/1.0"
+    
     for model in _MODELS:
-        for attempt in range(4):
+        for attempt in range(5):
             try:
-                fmt = prompt
                 resp = _req.post(
                     _FREE_API.format(model=model),
                     headers=headers,
                     json={
-                        "inputs": fmt,
+                        "inputs": prompt,
                         "parameters": {
                             "max_new_tokens": max_tokens,
                             "temperature": 0.7,
                             "return_full_text": False,
                             "do_sample": True,
                         },
-                        "options": {"wait_for_model": True}
+                        "options": {"wait_for_model": True, "use_cache": False}
                     },
                     timeout=180
                 )
@@ -61,19 +68,30 @@ def call_llama(prompt: str, max_tokens: int = 2000) -> str:
                         text = result[0].get("generated_text", "").strip()
                         if text:
                             return text
-                elif resp.status_code == 503:
-                    time.sleep(30 * (attempt + 1))
+                    elif isinstance(result, dict) and "generated_text" in result:
+                        text = result.get("generated_text", "").strip()
+                        if text:
+                            return text
+                
+                # 503 = model loading, retry with backoff
+                if resp.status_code == 503:
+                    wait_time = min(60 * (attempt + 1), 300)  # cap at 5 min
+                    time.sleep(wait_time)
                     continue
-                elif resp.status_code in [429, 402]:
+                
+                # 429/402 = rate limit or payment issue, skip model
+                if resp.status_code in [429, 402]:
                     break
-                else:
-                    time.sleep(10)
-                    continue
-            except Exception:
-                time.sleep(10)
+                
+                # Other errors, retry after delay
+                time.sleep(5 * (attempt + 1))
+                continue
+                
+            except Exception as e:
+                time.sleep(5 * (attempt + 1))
                 continue
     
-    return "Could not reach AI. Models may be loading — please try again in 30 seconds."
+    return "Could not reach AI models. Please check: 1) HF_TOKEN env var is set, 2) Network connection, 3) Hugging Face API status (try again in 60 seconds)."
 
 # =============================================================================
 # FILE EXTRACTION
@@ -490,4 +508,3 @@ if __name__ == "__main__":
         server_port=7860,
         show_error=True,
         share=False  # Spaces handles sharing automatically
-    )