support stream

app.py

@@ -205,23 +205,44 @@ def get_completion(message,history,profile: gr.OAuthProfile | None,oauth_token:
         'messages': messages,
         'temperature':temperature, 
         'max_tokens':max_tokens,
-        # 'stream':True，
+        'stream':True，
+        'stream_options':{"include_usage": True}, # retrieving token usage for stream response
     }
 
     # get response
-    response = requests.post('https://burn.hair/v1/chat/completions', headers=headers, json=data)
-    response_data = response.json()
-    print(response_data)
-    print('-----------------------------------\n')
-    if 'error' in response_data:
-        response_content = response_data['error']['message']
-    else:
-        response_content = response_data['choices'][0]['message']['content']
-        usage = response_data['usage']
-
+    # response = requests.post('https://burn.hair/v1/chat/completions', headers=headers, json=data)
+    # response_data = response.json()
+    # print(response_data)
+    # print('-----------------------------------\n')
+    # if 'error' in response_data:
+    #     response_content = response_data['error']['message']
+    # else:
+    #     response_content = response_data['choices'][0]['message']['content']
+    #     usage = response_data['usage']
+    # return response_content
+
+    # get response with stream
+    response = requests.post('https://burn.hair/v1/chat/completions', headers=headers, json=data,stream=True)
+    response_content = ""
+    for line in response.iter_lines():
+        line = line.decode().strip()
+        if line == "data: [DONE]":
+            continue
+        elif line.startswith("data: "):
+            line = line[6:] # remove prefix "data: "
+            try:
+                data = json.loads(line)
+                if "delta" in data["choices"][0]:
+                    content = data["choices"][0]["delta"].get("content", "")
+                    response_content += content
+                    yield response_content
+            except json.JSONDecodeError:
+                print(f"Error decoding line: {line}")
+
+    response_data = {}
+    
     supabase_insert_message(user_message,response_content,messages,response_data,user_name,user_oauth_token,ip,sign,cookie_value)
     
-    return response_content