Refactor GPTAssistantAgent (#632)

* Refactor GPTAssistantAgent constructor to handle
instructions and overwrite_instructions flag

- Ensure that `system_message` is always consistent with `instructions`
- Ensure provided instructions are always used
- Add option to permanently modify the instructions of the assistant

* Improve default behavior

* Add a test; add method to delete assistant

* Add a new test for overwriting instructions

* Add test case for when no instructions are given for existing assistant

* Add pytest markers to test_gpt_assistant.py

* add test in workflow

* update

* fix test_client_stream

* comment out test_hierarchy_

---------

Co-authored-by: Chi Wang <wang.chi@microsoft.com>
Co-authored-by: kevin666aa <yrwu000627@gmail.com>

test/agentchat/contrib/test_gpt_assistant.py

@@ -38,9 +38,10 @@ def test_gpt_assistant_chat():
         "description": "This is an API endpoint allowing users (analysts) to input question about GitHub in text format to retrieve the realted and structured data.",
     }
 
+    config_list = autogen.config_list_from_json(OAI_CONFIG_LIST, file_location=KEY_LOC)
     analyst = GPTAssistantAgent(
         name="Open_Source_Project_Analyst",
-        llm_config={"tools": [{"type": "function", "function": ossinsight_api_schema}]},
+        llm_config={"tools": [{"type": "function", "function": ossinsight_api_schema}], "config_list": config_list},
         instructions="Hello, Open Source Project Analyst. You'll conduct comprehensive evaluations of open source projects or organizations on the GitHub platform",
     )
     analyst.register_function(
@@ -62,5 +63,114 @@ def test_gpt_assistant_chat():
     assert len(analyst._openai_threads) == 0
 
 
+@pytest.mark.skipif(
+    sys.platform in ["darwin", "win32"] or skip_test,
+    reason="do not run on MacOS or windows or dependency is not installed",
+)
+def test_get_assistant_instructions():
+    """
+    Test function to create a new GPTAssistantAgent, set its instructions, retrieve the instructions,
+    and assert that the retrieved instructions match the set instructions.
+    """
+
+    config_list = autogen.config_list_from_json(OAI_CONFIG_LIST, file_location=KEY_LOC)
+    assistant = GPTAssistantAgent(
+        "assistant",
+        instructions="This is a test",
+        llm_config={
+            "config_list": config_list,
+        },
+    )
+
+    instruction_match = assistant.get_assistant_instructions() == "This is a test"
+    assistant.delete_assistant()
+
+    assert instruction_match is True
+
+
+@pytest.mark.skipif(
+    sys.platform in ["darwin", "win32"] or skip_test,
+    reason="do not run on MacOS or windows or dependency is not installed",
+)
+def test_gpt_assistant_instructions_overwrite():
+    """
+    Test that the instructions of a GPTAssistantAgent can be overwritten or not depending on the value of the
+    `overwrite_instructions` parameter when creating a new assistant with the same ID.
+
+    Steps:
+    1. Create a new GPTAssistantAgent with some instructions.
+    2. Get the ID of the assistant.
+    3. Create a new GPTAssistantAgent with the same ID but different instructions and `overwrite_instructions=True`.
+    4. Check that the instructions of the assistant have been overwritten with the new ones.
+    """
+
+    instructions1 = "This is a test #1"
+    instructions2 = "This is a test #2"
+
+    config_list = autogen.config_list_from_json(OAI_CONFIG_LIST, file_location=KEY_LOC)
+    assistant = GPTAssistantAgent(
+        "assistant",
+        instructions=instructions1,
+        llm_config={
+            "config_list": config_list,
+        },
+    )
+
+    assistant_id = assistant.assistant_id
+    assistant = GPTAssistantAgent(
+        "assistant",
+        instructions=instructions2,
+        llm_config={
+            "config_list": config_list,
+            "assistant_id": assistant_id,
+        },
+        overwrite_instructions=True,
+    )
+
+    instruction_match = assistant.get_assistant_instructions() == instructions2
+    assistant.delete_assistant()
+
+    assert instruction_match is True
+
+
+@pytest.mark.skipif(
+    sys.platform in ["darwin", "win32"] or skip_test,
+    reason="do not run on MacOS or windows or dependency is not installed",
+)
+def test_gpt_assistant_existing_no_instructions():
+    """
+    Test function to check if the GPTAssistantAgent can retrieve instructions for an existing assistant
+    even if the assistant was created with no instructions initially.
+    """
+    instructions = "This is a test #1"
+
+    config_list = autogen.config_list_from_json(OAI_CONFIG_LIST, file_location=KEY_LOC)
+    assistant = GPTAssistantAgent(
+        "assistant",
+        instructions=instructions,
+        llm_config={
+            "config_list": config_list,
+        },
+    )
+
+    assistant_id = assistant.assistant_id
+
+    # create a new assistant with the same ID but no instructions
+    assistant = GPTAssistantAgent(
+        "assistant",
+        llm_config={
+            "config_list": config_list,
+            "assistant_id": assistant_id,
+        },
+    )
+
+    instruction_match = assistant.get_assistant_instructions() == instructions
+    assistant.delete_assistant()
+    assert instruction_match is True
+
+
 if __name__ == "__main__":
     test_gpt_assistant_chat()
+    test_get_assistant_instructions()
+    test_gpt_assistant_instructions_overwrite()
+    test_gpt_assistant_existing_no_instructions()

test/oai/test_client_stream.py

@@ -18,7 +18,7 @@ def test_aoai_chat_completion_stream():
         filter_dict={"api_type": ["azure"], "model": ["gpt-3.5-turbo"]},
     )
     client = OpenAIWrapper(config_list=config_list)
-    response = client.create(messages=[{"role": "user", "content": "2+2="}], seed=None, stream=True)
+    response = client.create(messages=[{"role": "user", "content": "2+2="}], stream=True)
     print(response)
     print(client.extract_text_or_function_call(response))
 
@@ -31,7 +31,7 @@ def test_chat_completion_stream():
         filter_dict={"model": ["gpt-3.5-turbo"]},
     )
     client = OpenAIWrapper(config_list=config_list)
-    response = client.create(messages=[{"role": "user", "content": "1+1="}], seed=None, stream=True)
+    response = client.create(messages=[{"role": "user", "content": "1+1="}], stream=True)
     print(response)
     print(client.extract_text_or_function_call(response))
 
@@ -63,7 +63,6 @@ def test_chat_functions_stream():
     response = client.create(
         messages=[{"role": "user", "content": "What's the weather like today in San Francisco?"}],
         functions=functions,
-        seed=None,
         stream=True,
     )
     print(response)
@@ -74,7 +73,7 @@ def test_chat_functions_stream():
 def test_completion_stream():
     config_list = config_list_openai_aoai(KEY_LOC)
     client = OpenAIWrapper(config_list=config_list)
-    response = client.create(prompt="1+1=", model="gpt-3.5-turbo-instruct", seed=None, stream=True)
+    response = client.create(prompt="1+1=", model="gpt-3.5-turbo-instruct", stream=True)
     print(response)
     print(client.extract_text_or_function_call(response))
 

test/test_notebook.py

@@ -84,12 +84,12 @@ def _test_oai_chatgpt_gpt4(save=False):
     run_notebook("oai_chatgpt_gpt4.ipynb", save=save)
 
 
-@pytest.mark.skipif(
-    skip or not sys.version.startswith("3.10"),
-    reason="do not run if openai is not installed or py!=3.10",
-)
-def test_hierarchy_flow_using_select_speaker(save=False):
-    run_notebook("agentchat_hierarchy_flow_using_select_speaker.ipynb", save=save)
+# @pytest.mark.skipif(
+#     skip or not sys.version.startswith("3.10"),
+#     reason="do not run if openai is not installed or py!=3.10",
+# )
+# def test_hierarchy_flow_using_select_speaker(save=False):
+#     run_notebook("agentchat_hierarchy_flow_using_select_speaker.ipynb", save=save)
 
 
 if __name__ == "__main__":