Update api.py

api.py

@@ -10,47 +10,33 @@ from deep_translator import GoogleTranslator
 
 # ---------- Fix OpenAI version issue ----------
 try:
-    # Ensure modern OpenAI SDK is installed and loaded
     subprocess.run(
         ["pip", "install", "--no-cache-dir", "--upgrade", "openai==1.12.0"],
         check=True,
     )
     import openai
     importlib.reload(openai)
-    from openai import OpenAI
     print("✅ Using OpenAI SDK version 1.12.0")
 except Exception as e:
     print("⚠️ Could not ensure correct OpenAI version:", e)
-    from openai import OpenAI  # fallback import
+    import openai
 
-# Patch huggingface_hub automatically if Gradio overwrote it
+# ---------- Patch huggingface_hub dynamically ----------
 try:
     import huggingface_hub
     if not hasattr(huggingface_hub, "cached_download"):
-        subprocess.run(
-            [
-                "pip",
-                "install",
-                "--no-cache-dir",
-                "huggingface-hub==0.24.5",
-                "transformers==4.30.2",
-                "sentence-transformers==2.2.2",
-            ],
-            check=True,
-        )
-        print("✅ Downgraded huggingface-hub for sentence-transformers compatibility.")
+        print("⚙️ Adding compatibility patch for huggingface_hub.cached_download()")
+        huggingface_hub.cached_download = huggingface_hub.hf_hub_download
 except Exception as e:
-    print("⚠️ Could not auto-patch huggingface_hub:", e)
+    print("⚠️ Could not patch huggingface_hub:", e)
 
 from sentence_transformers import SentenceTransformer
 from pinecone import Pinecone
 
 # ---------- Config ----------
 DATASET_PATH = "data/coaching_millionaer_dataset.json"
-# Load .env (for local dev), but also check Hugging Face environment
 load_dotenv(override=True)
 
-# Ensure environment variables are loaded even if running on Hugging Face
 OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") or os.environ.get("OPENAI_API_KEY")
 PINECONE_API_KEY = os.getenv("PINECONE_API_KEY") or os.environ.get("PINECONE_API_KEY")
 
@@ -61,29 +47,22 @@ app = Flask(__name__)
 CORS(app, resources={r"/ask": {"origins": "*"}})
 
 # ---------- OpenAI Client ----------
-import openai
-
 client = None
 try:
     if not OPENAI_API_KEY or OPENAI_API_KEY.strip() == "":
         raise ValueError("Missing OPENAI_API_KEY in environment variables")
 
-    # Handle both old and new SDK structures safely
-    try:
-        # New SDK style
-        client = openai.OpenAI(api_key=OPENAI_API_KEY)
-        print("✅ Using new OpenAI client interface.")
-    except Exception:
-        # Fallback to legacy interface
-        openai.api_key = OPENAI_API_KEY
-        client = openai
-        print("✅ Using legacy OpenAI interface.")
+    # Always use global API style (avoids proxies bug)
+    import openai
+    openai.api_key = OPENAI_API_KEY
+    client = openai
+    print("✅ OpenAI client initialized using legacy API style.")
+    print(f"✅ Using OpenAI SDK version {openai.__version__}")
 
 except Exception as e:
     client = None
     print(f"❌ Failed to initialize OpenAI client: {e}")
 
-
 # ---------- Retriever ----------
 retriever = None
 try:
@@ -123,7 +102,6 @@ except Exception as e:
 
 # ---------- Translator ----------
 def translate_text(text: str, target_lang: str) -> str:
-    """Translate text using deep-translator (GoogleTranslator)."""
     try:
         return GoogleTranslator(source="auto", target=target_lang).translate(text)
     except Exception:
@@ -131,14 +109,12 @@ def translate_text(text: str, target_lang: str) -> str:
 
 # ---------- Helpers ----------
 def detect_language(question: str) -> str:
-    """Detect the user's language without translation."""
     try:
         return detect(question)
     except Exception:
         return "unknown"
 
 def normalize_language(lang: str, text: str) -> str:
-    """Fix incorrect language detection like 'wer is' → German."""
     if lang == "nl" and any(
         word in text.lower() for word in ["wer", "was", "wie", "javid", "coaching"]
     ):
@@ -202,15 +178,13 @@ def ask():
 
     print(f"\n--- User Question ---\n{question}")
 
-    # Detect and normalize language
     user_lang = normalize_language(detect_language(question), question)
     print(f"Detected language: {user_lang}")
 
-    # Retrieve context
     context, results = "", []
     try:
         raw_results = retriever.retrieve(question)
-        MIN_SCORE = 0.10  # Pinecone similarity scores are normalized (0–1)
+        MIN_SCORE = 0.10
         results = [r for r in raw_results if r.get("score", 0) >= MIN_SCORE]
         if results:
             context = "\n\n---\n\n".join(
@@ -220,7 +194,6 @@ def ask():
         traceback.print_exc()
         return jsonify(format_answers(question, f"Retriever error: {e}", [])), 200
 
-    # Build prompts
     if context:
         sys_prompt = system_prompt_book_only()
         user_content = f"Question: {question}\n\nBook context:\n{context}"
@@ -228,17 +201,31 @@ def ask():
         sys_prompt = system_prompt_fallback()
         user_content = question
 
-    # Query GPT
+    # ---------- OpenAI Query ----------
     try:
-        response = client.chat.completions.create(
-            model="gpt-4o-mini",
-            messages=[
-                {"role": "system", "content": sys_prompt},
-                {"role": "user", "content": user_content},
-            ],
-            max_tokens=700,
-        )
-        answer = response.choices[0].message.content.strip()
+        if hasattr(client, "ChatCompletion"):
+            # Legacy API (works in SDK <=1.12.0)
+            response = client.ChatCompletion.create(
+                model="gpt-4o-mini",
+                messages=[
+                    {"role": "system", "content": sys_prompt},
+                    {"role": "user", "content": user_content},
+                ],
+                max_tokens=700,
+            )
+            answer = response["choices"][0]["message"]["content"].strip()
+        else:
+            # New API (for future SDKs)
+            response = client.chat.completions.create(
+                model="gpt-4o-mini",
+                messages=[
+                    {"role": "system", "content": sys_prompt},
+                    {"role": "user", "content": user_content},
+                ],
+                max_tokens=700,
+            )
+            answer = response.choices[0].message.content.strip()
+
     except Exception as e:
         traceback.print_exc()
         return jsonify(format_answers(question, f"⚠️ OpenAI call failed: {e}", [])), 200