Fix memory step model output in ToolCallingAgent #1156
Conversation
|
Nice, thank you! Could you also test the |
|
I tried to add the test like shown below agent = ToolCallingAgent(tools=[PythonInterpreterTool()], model=FakeToolCallModel(), verbosity_level=0)
agent.run("What is 2 multiplied by 3.6452?")
with agent.logger.console.capture() as capture:
agent.replay()
output = capture.get().replace("\n", "")
assert "Called Tool" in outputbut captured output includes formatting codes. |
|
@keetrap I see string |
| @@ -160,7 +160,7 @@ def test_streaming_agent_image_output(self): | |||
| ) | |||
| ) | |||
|
|
|||
| self.assertEqual(len(outputs), 5) | |||
| self.assertEqual(len(outputs), 6) | |||
There was a problem hiding this comment.
We have one extra ChatMessage in outputs here.
In this case it is ChatMessage(content="Called Tool: 'final_answer' with arguments: {'answer': 'image'}", role='assistant', metadata={}, options=[])
``
|
You can either remove ansi sequences before asserting, something like: def remove_ansi(text):
ansi_escape = re.compile(r'\x1B\[[0-?]*[ -/]*[@-~]')
return ansi_escape.sub('', text)or just disable them altogether by changing AgentLogger: logger = AgentLogger(2)
logger.console = Console(force_terminal=False)
# inject or monkeypatch the loggerActually I have no idea why |
|
@sysradium agree that being able to pass a console arg upon logger initialization would be better! If you have some time, it'd be great to make a PR for this! |
Fixes #1143