Switch to current version of OpenAI.

Introduces several changes with the main objective of enabling OpenAI's streaming API in the chatmastermind application. This allows for the retrieval of AI responses gradually as a stream, which can significantly improve the user experience in interactions that involve large result sets.

* Added tiktoken import in 'openai.py' and modifications to the OpenAI class to support streaming. This includes the addition of a new class OpenAIAnswer to handle streaming API responses.
* Modified request function in the OpenAI class: the stream=True flag is added to the openai.ChatCompletion.create method to enable streaming API.
* Modified 'question.py' to print the answer parts as they are streamed.
* Replaced the Answer class's string data type with a generator which supports str and Generator[str, None, None] data types. Modifications are made to the Answer class methods to handle both data types accordingly.
* Updated the tests in 'test_ais_openai.py' and 'test_message.py' to reflect and validate these changes.

chatmastermind/ais/openai.py

@@ -47,12 +47,12 @@ class OpenAIAnswer:
             self.finished = True
         if not self.finished:
             found_choice = False
-            for choice in chunk['choices']:
-                if not choice['finish_reason']:
-                    self.streams[choice['index']].data.append(choice['delta']['content'])
-                    self.tokens.completion += len(self.encoding.encode(choice['delta']['content']))
+            for choice in chunk.choices:
+                if not choice.finish_reason:
+                    self.streams[choice.index].data.append(choice.delta.content)
+                    self.tokens.completion += len(self.encoding.encode(choice.delta.content))
                     self.tokens.total = self.tokens.prompt + self.tokens.completion
-                if choice['index'] == self.idx:
+                if choice.index == self.idx:
                     found_choice = True
             if not found_choice:
                 return False
@@ -68,6 +68,10 @@ class OpenAI(AI):
         self.ID = config.ID
         self.name = config.name
         self.config = config
+        self.client = openai.OpenAI(api_key=self.config.api_key)
+
+    def _completions(self, *args, **kw):  # type: ignore
+        return self.client.chat.completions.create(*args, **kw)
 
     def request(self,
                 question: Message,
@@ -80,10 +84,9 @@ class OpenAI(AI):
         nr. of messages in the 'AIResponse'.
         """
         self.encoding = tiktoken.encoding_for_model(self.config.model)
-        openai.api_key = self.config.api_key
         oai_chat, prompt_tokens = self.openai_chat(chat, self.config.system, question)
         tokens: Tokens = Tokens(prompt_tokens, 0, prompt_tokens)
-        response = openai.ChatCompletion.create(
+        response = self._completions(
             model=self.config.model,
             messages=oai_chat,
             temperature=self.config.temperature,
@@ -114,9 +117,8 @@ class OpenAI(AI):
         Return all models supported by this AI.
         """
         ret = []
-        for engine in sorted(openai.Engine.list()['data'], key=lambda x: x['id']):
-            if engine['ready']:
-                ret.append(engine['id'])
+        for engine in sorted(self.client.models.list().data, key=lambda x: x.id):
+            ret.append(engine.id)
         ret.sort()
         return ret
 
@@ -124,14 +126,8 @@ class OpenAI(AI):
         """
         Print all models supported by the current AI.
         """
-        not_ready = []
-        for engine in sorted(openai.Engine.list()['data'], key=lambda x: x['id']):
-            if engine['ready']:
-                print(engine['id'])
-            else:
-                not_ready.append(engine['id'])
-        if len(not_ready) > 0:
-            print('\nNot ready: ' + ', '.join(not_ready))
+        for model in self.models():
+            print(model)
 
     def openai_chat(self, chat: Chat, system: str,
                     question: Optional[Message] = None) -> tuple[ChatType, int]:
@@ -150,7 +146,7 @@ class OpenAI(AI):
         for message in chat.messages:
             if message.answer:
                 prompt_tokens += append('user', message.question)
-                prompt_tokens += append('assistant', message.answer)
+                prompt_tokens += append('assistant', str(message.answer))
         if question:
             prompt_tokens += append('user', question.question)
         return oai_chat, prompt_tokens

tests/test_ais_openai.py

@@ -9,43 +9,31 @@ from chatmastermind.configuration import OpenAIConfig
 
 class OpenAITest(unittest.TestCase):
 
-    @mock.patch('openai.ChatCompletion.create')
+    @mock.patch('chatmastermind.ais.openai.OpenAI._completions')
     def test_request(self, mock_create: mock.MagicMock) -> None:
         # Create a test instance of OpenAI
         config = OpenAIConfig()
         openai = OpenAI(config)
 
         # Set up the mock response from openai.ChatCompletion.create
-        mock_chunk1 = {
-            'choices': [
-                {
-                    'index': 0,
-                    'delta': {
-                        'content': 'Answer 1'
-                    },
-                    'finish_reason': None
-                },
-                {
-                    'index': 1,
-                    'delta': {
-                        'content': 'Answer 2'
-                    },
-                    'finish_reason': None
-                }
-            ],
-        }
-        mock_chunk2 = {
-            'choices': [
-                {
-                    'index': 0,
-                    'finish_reason': 'stop'
-                },
-                {
-                    'index': 1,
-                    'finish_reason': 'stop'
-                }
-            ],
-        }
+        class mock_obj:
+            pass
+        mock_chunk1 = mock_obj()
+        mock_chunk1.choices = [mock_obj(), mock_obj()]  # type: ignore
+        mock_chunk1.choices[0].index = 0  # type: ignore
+        mock_chunk1.choices[0].delta = mock_obj()  # type: ignore
+        mock_chunk1.choices[0].delta.content = 'Answer 1'  # type: ignore
+        mock_chunk1.choices[0].finish_reason = None  # type: ignore
+        mock_chunk1.choices[1].index = 1  # type: ignore
+        mock_chunk1.choices[1].delta = mock_obj()  # type: ignore
+        mock_chunk1.choices[1].delta.content = 'Answer 2'  # type: ignore
+        mock_chunk1.choices[1].finish_reason = None  # type: ignore
+        mock_chunk2 = mock_obj()
+        mock_chunk2.choices = [mock_obj(), mock_obj()]  # type: ignore
+        mock_chunk2.choices[0].index = 0  # type: ignore
+        mock_chunk2.choices[0].finish_reason = 'stop'  # type: ignore
+        mock_chunk2.choices[1].index = 1  # type: ignore
+        mock_chunk2.choices[1].finish_reason = 'stop'  # type: ignore
         mock_create.return_value = iter([mock_chunk1, mock_chunk2])
 
         # Create test data