Chitchat cannot be handled

Rasa Pro CALM
1

Hi. I’m trying to enable chitchat in my Rasa Pro CALM bot. I did exactly what is written here: Chitchat and Small Talk , but after any chitchat message, I get the default “cannot handle” message as a response.

I have

  • fully functioning Rasa CALM where everything else works well
  • Intentess policy set up in config.yml
  • rephrase turned on in endpoints.yml
  • working OpenAI key in the env

In the logs I see the Chitchat() command is correctly predicted, but a bit later “cannot_handle” pattern is triggered with reason “cannot_handle_chitchat”. Here is what I think is the relevant part of the logs:

2025-02-13 14:10:00 DEBUG    rasa.shared.providers._ssl_verification_utils  - [debug    ] ensure_ssl_certificates_for_litellm_non_openai_based_clients ssl_certificate=None ssl_verify=None
2025-02-13 14:10:00 DEBUG    rasa.shared.providers._ssl_verification_utils  - [debug    ] ensure_ssl_certificates_for_litellm_openai_based_clients ssl_certificate=None ssl_verify=None
2025-02-13 14:10:00 DEBUG    rasa.shared.providers.llm._base_litellm_client  - [debug    ] base_litellm_client.formatted_response formatted_response={'id': 'chatcmpl-B0SktHKKj9ujsDzIK1AXhZg0nTzLA', 'choices': ['ChitChat()'], 'created': 1739450235, 'model': 'gpt-4-0613', 'usage': {'prompt_tokens': 730, 'completion_tokens': 5, 'total_tokens': 735}, 'additional_info': None}
2025-02-13 14:10:00 DEBUG    rasa.utils.log_utils  - [debug    ] llm_command_generator.predict_commands.actions_generated action_list=ChitChat()
C:\Users\PetrG\AppData\Local\miniconda3\envs\Chatbot-assistant\lib\site-packages\pydantic\main.py:314: UserWarning: Pydantic serializer warnings:
  Expected `CompletionTokensDetails` but got `dict` - serialized value may not be as expected
  return self.__pydantic_serializer__.to_python(
2025-02-13 14:10:00 DEBUG    rasa.utils.log_utils  - [debug    ] llm_command_generator.predict_commands.finished commands=[ChitChatAnswerCommand()]
2025-02-13 14:10:00 DEBUG    rasa.engine.graph  - [debug    ] graph.node.running_component   clazz=RegexMessageHandler fn=process node_name=run_RegexMessageHandler
2025-02-13 14:10:00 DEBUG    rasa.core.processor  - [debug    ] processor.message.parse        parse_data_entities=[] parse_data_intent={'name': None, 'confidence': 0.0} parse_data_text=how are you?
2025-02-13 14:10:00 DEBUG    rasa.core.processor  - Logged UserUtterance - tracker now has 4 events.
2025-02-13 14:10:00 DEBUG    rasa.engine.runner.dask  - Running graph with inputs: {'__tracker__': DialogueStateTracker(sender_id: a4870e22aff94850840a49e6947c7018)}, targets: ['command_processor'] and ExecutionContext(model_id='f4e31ef37a464f6b83d396a9ec864be2', should_add_diagnostic_data=False, is_finetuning=False, node_name=None).
2025-02-13 14:10:00 DEBUG    rasa.engine.graph  - [debug    ] graph.node.running_component   clazz=FlowsProvider fn=provide_inference node_name=flows_provider
2025-02-13 14:10:00 DEBUG    rasa.engine.graph  - [debug    ] graph.node.running_component   clazz=StoryGraphProvider fn=provide_inference node_name=story_graph_provider
2025-02-13 14:10:00 DEBUG    rasa.engine.graph  - [debug    ] graph.node.running_component   clazz=CommandProcessorComponent fn=execute_commands node_name=command_processor
2025-02-13 14:10:00 WARNING  rasa.dialogue_understanding.processor.command_processor  - [warning  ] command_processor.clean_up_chitchat_command.replace_chitchat_answer_with_cannot_handle command=CannotHandleCommand(reason='cannot_handle_chitchat') defined_intentless_policy_in_config=True pattern_chitchat_uses_action_trigger_chitchat=True
2025-02-13 14:10:00 DEBUG    rasa.dialogue_understanding.processor.command_processor  - [debug    ] command_processor.clean_up_commands.final_commands command=[CannotHandleCommand(reason='cannot_handle_chitchat')]
2025-02-13 14:10:00 DEBUG    rasa.engine.runner.dask  - Running graph with inputs: {'__tracker__': DialogueStateTracker(sender_id: a4870e22aff94850840a49e6947c7018), '__endpoints__': <rasa.core.utils.AvailableEndpoints object at 0x000002097B10CE50>}, targets: ['select_prediction'] and ExecutionContext(model_id='f4e31ef37a464f6b83d396a9ec864be2', should_add_diagnostic_data=False, is_finetuning=False, node_name=None).

No idea what’s wrong and how to make it work…

Thanks much for any help.

2 Likes
2

According to info here - Conversation Repair :

** The pattern pattern_cannot_handle has the context attribute reason. reason is set to:

  • cannot_handle_chitchat, when pattern_chitchat attempts to invoke action_trigger_chitchat without an IntentlessPolicy defined.**

But I have IntentlessPolicy in my config.yml and the logs from rasa train suggest it is correctly trained:

2025-02-13 14:54:06 INFO     rasa.engine.training.hooks  - Starting to train component 'IntentlessPolicy'.
2025-02-13 14:54:07 INFO     rasa.core.policies.intentless_policy  - [info     ] intentless_policy.training.completed
2025-02-13 14:54:08 INFO     rasa.engine.training.hooks  - Finished training component 'IntentlessPolicy'.
2 Likes
3

Commenting to share that I am facing the exact same issue after following the documentation. Have you found a solution yet or is this still an issue?

I am limited to using Rasa 3.10 but would good to know if it’s also an issue with later versions.

2025-02-19 15:13:55 WARNING  rasa.dialogue_understanding.processor.command_processor  - [warning  ] command_processor.clean_up_chitchat_command.replace_chitchat_answer_with_cannot_
handle command=CannotHandleCommand(reason='cannot_handle_chitchat') defined_intentless_policy_in_config=True pattern_chitchat_uses_action_trigger_chitchat=True
1 Like
4

Nope. I tried everything that I thought might be relevant - no success. Also couldn’t find any relevant info on this anywhere else. I’m on Rasa 3.10.7.

Currently I have a workaround, where I use a “chitchat” flow instead - this handles at least the most frequent cases of user chitchat messages in a simple manner.

Would be grateful if you let me know in case you solve this.

1 Like
5

@Arjaan can you please advise when we can expect support on this issue? This thread is now a week old and following the official Rasa documentation has been insufficient.

2 Likes
8

Do you found how to fix it ? :grimacing:

9

Nope. Could please someone from Rasa comment on this? I’m clearly not the only one having this problem and the documentation is not sufficient on this.

10

@PetrG I tried tagging the person who wrote the pinned post saying Rasa devs monitor these threads daily. I suggest tagging others if you know any

11

@ChrisRahme : Would you perhaps know how to solve this…? Looks like more people face the same issue and we’re not getting any help here.. Thanks..

12

Hi all, thank you for raising this issue and sorry for making you wait, I’ll check with the Rasa team what’s the best way to solve it and come back to you shortly, Marina

13

Hi @PetrG @mnchs12312321 @shellka thanks again for raising this issue. This is a bug and it has been reported.

This bug is triggered when IntentlessPolicy is used AND there is no e2e tests. If you want to continue using IntentlessPolicy there is a workaround: please create an e2e test, here is the link to docs how to do it: Evaluating Your Assistant (E2E Testing) | Rasa Documentation.

Please let me know if you still encounter the issue. Marina

1 Like
14

I am still having the same issue here.