Update app.py

app.py

@@ -49,11 +49,11 @@ def convert_to_openai_format(azure_response):
     
     return {k: v for k, v in openai_response.items() if v is not None}
 
-@app.route('/hf/v1/chat/completions', methods=['POST'])
+@app.route('/v1/chat/completions', methods=['POST'])
 @token_required
 def chat_completions():
-    openai_data = request.json
-    stream = openai_data.get('stream', False)
+    request_data = request.json
+    stream = request_data.get('stream', False)
 
     azure_url = f"{AZURE_API_BASE}/openai/deployments/{AZURE_DEPLOYMENT_NAME}/chat/completions?api-version={AZURE_API_VERSION}"
 
@@ -63,7 +63,7 @@ def chat_completions():
     }
 
     if not stream:
-        response = requests.post(azure_url, json=openai_data, headers=headers)
+        response = requests.post(azure_url, json=request_data, headers=headers)
 
         if response.status_code == 200:
             azure_response = response.json()
@@ -73,7 +73,7 @@ def chat_completions():
             return jsonify({"error": "Azure OpenAI API request failed", "details": response.text}), response.status_code
     else:
         def generate():
-            with requests.post(azure_url, json=openai_data, headers=headers, stream=True) as response:
+            with requests.post(azure_url, json=request_data, headers=headers, stream=True) as response:
                 if response.status_code != 200:
                     yield f"data: {json.dumps({'error': 'Azure OpenAI API request failed', 'details': response.text})}\n\n"
                     return