Update app.py

app.py

@@ -4,7 +4,6 @@ from fastapi.responses import StreamingResponse, JSONResponse
 from pydantic import BaseModel
 from typing import List, Optional
 
-import asyncio
 import json
 import time
 import uuid
@@ -35,7 +34,7 @@ API_KEY = "sk-your-secret-key"
 
 app = FastAPI(
     title="Universal AI Gateway",
-    version="5.2.0"
+    version="5.3.0"
 )
 
 
@@ -77,6 +76,7 @@ def verify_api_key(req: Request):
 
     auth = req.headers.get("Authorization")
 
+    # السماح بدون مفتاح للاختبار
     if not auth:
         return True
 
@@ -107,7 +107,7 @@ async def root():
     return {
         "status": "online",
         "service": "Universal AI Gateway",
-        "version": "5.2.0"
+        "version": "5.3.0"
     }
 
 
@@ -145,40 +145,6 @@ async def get_models():
     }
 
 
-# =====================================================
-# NORMAL COMPLETION
-# =====================================================
-
-def normal_completion(model, messages):
-
-    client = Client()
-
-    response = client.chat.completions.create(
-        model=model,
-        messages=messages,
-        stream=False
-    )
-
-    return response
-
-
-# =====================================================
-# STREAM COMPLETION
-# =====================================================
-
-def stream_completion(model, messages):
-
-    client = Client()
-
-    response = client.chat.completions.create(
-        model=model,
-        messages=messages,
-        stream=True
-    )
-
-    return response
-
-
 # =====================================================
 # CHAT COMPLETIONS
 # =====================================================
@@ -209,16 +175,18 @@ async def chat_completions(
 
     if body.stream:
 
-        async def generate_stream():
+        def generate_stream():
 
             chunk_id = f"chatcmpl-{uuid.uuid4().hex}"
 
             try:
 
-                response = await asyncio.to_thread(
-                    stream_completion,
-                    body.model,
-                    messages
+                client = Client()
+
+                response = client.chat.completions.create(
+                    model=body.model,
+                    messages=messages,
+                    stream=True
                 )
 
                 has_content = False
@@ -232,15 +200,12 @@ async def chat_completions(
                         if (
                             hasattr(chunk, "choices")
                             and chunk.choices
+                            and len(chunk.choices) > 0
+                            and hasattr(chunk.choices[0], "delta")
                             and chunk.choices[0].delta
                             and chunk.choices[0].delta.content
                         ):
-                            content = (
-                                chunk
-                                .choices[0]
-                                .delta
-                                .content
-                            )
+                            content = chunk.choices[0].delta.content
 
                         if not content:
                             continue
@@ -352,10 +317,12 @@ async def chat_completions(
 
     try:
 
-        response = await asyncio.to_thread(
-            normal_completion,
-            body.model,
-            messages
+        client = Client()
+
+        response = client.chat.completions.create(
+            model=body.model,
+            messages=messages,
+            stream=False
         )
 
         assistant_message = ""