CORS error for server requests [SOLVED]

This worked for me.

Any way to disable info messages in Rasa shell

So I sometimes use rasa shell to quickly test out my model, because I print a lot of debug statements from the custom actions. I have now moved to Rasa Pro 3.10.8 and wanted to use the direct custom actions, where I run the actions python file directly within the rasa assistant. However, for every …

Local llm error with enviromental variable OPENAI key

Can you try without mentioning the type as openai, and also without the openai_api_key? Like it is done here:

RASA not working with ollama

I get the same warning with Ollama, and the response times are so bad that its quicker to use GPT-4. But what API do you mean here exactly?

How to choose response based on flow

Can you share the link for the appropriate section of the documentation, or an example?

Thank you. I somehow overlooked this in the documentation. This answers my question perfectly.

I would like to choose from the default utter_responses in the domain file, based on the current active flow. The use case is that I have a common slot between multiple flows, and I want to change the utter_ask_slot response based on the current active flow. If possible, I would like to do it without…

S exceeded its OpenAI usage limits, causing problems during rasa pro training (even though I'm using gemini api)

Maybe try setting the flow_retrieval under the SingleStepLLMCommandGenerator to false. This uses OpenAI embeddings by default and might be causing this issue.

- name: SingleStepLLMCommandGenerator
  llm:
    provider: gemini
    model: "gemini-1.5-flash"
    flow_retrieval:
      active: false

How to override slots in triggered flows

So in my use case, consider I said

Move the cup

Cup is extracted in the Object slot and the chatbot asks me for a destination. If I enter a new instruction in an attempt to override the object, like say

Move the bottle

I would expect it to override the object, but instead, I get pattern_cann…

Modify pattern_clarification

I have like super detailed flow descriptions that are a couple of lines themselves. I even added examples. My main issue is that I want this to be working with smaller LLMs. When using GPT-4o I got no issues. But something like phi-3 or gemma2 and it really struggles when the user input is a bit ambiguous.

I would like to use the pattern_classification, but modify it to instead use an LLM call to paraphrase the user utterance. I can get this part done using a custom action, but what I would like to do is to actually use the paraphrased sentence to continue the command generation. To make it more clear…

How to use NLU entities in CALM flows

Hi, first off, yes it is not really natural language input. I just want to trigger the flows from another process that is running elsewhere, but I was unable to get it working by directly triggering the intents via API (Open forum post here) so this is my workaround for that. As for having it as bu…

I have the following as my config, where the NLU part is based on some example from the rasa calm demo github repo

pipeline:
  - name: WhitespaceTokenizer
  - name: CountVectorsFeaturizer
    analyzer: char_wb
    min_ngram: 1
    max_ngram: 4
  - name: LogisticRegression

Minimal viable LLM for Command Generation?

I personally used phi3:14b-instruct from Ollama and it worked decently well. However, a point I noticed is that if you have multiple similar flows with only minor changes, the smaller models fail to capture the differences. But if it's for entirely different tasks or conversations, Phi3 worked well for…

Gpt-35-turbo with RASA CALM error

Yes, it is supported. I am using it with gpt-3.5-turbo-0125 Maybe share what your error is, as well as the config file

How to use Rasa Inspect with a rasa run server

So I need to use rasa run --enable-api because I need multiple instances sending input to Rasa, but I also need a primary interface, and I would very much prefer it was the Rasa inspect interface. So basically I interact from the inspect interface chat window, but I want the custom actions to trigger…

Does someone try LLMIntentClassifier?

As far as I know, LLMCommandGenerator is only available in Rasa Pro. So this is not really a question for Rasa open source.

Collect slots in one message

I haven’t exactly tried collecting so many varied slots all at once, but if the user did say all the values in the same sentence then the LLM should theoretically be able to fill in all the slots. So your first message could be asking to say all these details? But I wouldn’t recommend this approach,…

OpenAI package versioning

But I should assume I would still need different python environments?

I intend to use OpenAI package within my custom actions but Rasa-pro has its requirements set to a very old version, and I am wondering if me updating this package will break any functionality. rasa-pro 3.8.7 requires openai<0.29.0,>=0.28.1, but you have openai 1.35.9 which is incompatible.

How to access currently used LLM

How do I send custom chat completion requests to the currently used LLM from within a custom action? I constantly test out my Rasa CALM model with both gpt-4 and local LLMs on Ollama, and I want to send a request to whichever LLM server I have configured without having to change the code directly.

How to deal with flow interruptions

yes these two flows are two similar, and I would understand it if the LLM made a mistake in choosing between one of these. But what happened is that it assigned a value to the slot without asking anything. There was no mention of a bedroom in that conversation, and yet it assigned that in the source…

this is for when the user does not specify the source location

move_object_no_location:
  description: This flow is triggered when the user requests the robot to move an object to a destination without specifying its source location.
  steps:
    - collect: object
      description: Extracts the object from the user input.

Triggering flows with injected intents

I have defined a flow with the nlu_trigger and intent, and I have added also the NLUCommandAdapter. However, when I use the HTTP API to trigger the intent, the flow does not get activated, but I just receive the proper response for code 200. Am I doing it wrong or can we not use the trigger intent to…

So I am developing a rasa calm chatbot for interacting with a simulated robot of sorts. I ask it to move objects and once it has all the required slots, it will trigger a custom action. However, if I interrupt an ongoing flow, it behaves weirdly and sets slots automatically. Here is an example: user…

Question about Knowledge base for slots

That is my current approach, but I just wanted to know if there was any way I can get the LLM to do it for me.

From what I understand of the docs, the Knowledge base document is only used when pattern_search is triggered, and these are usually for answering questions. I was wondering if there was anyway we can use a DB as vector store and then use the values to set slots. Suppose, I say to the conversation a…

Ollama as embeddings provider

I understand that Ollama can be used as an LLM provider via the OpenAI compatible API, but how to utilize it as an embeddings provider. According to the blog, Ollama’s OpenAI API compatibility for embeddings is still in the works, and since huggingface_hub has an open issue right now where it cannot…

Where are predefined responses located

This is the log from when I run rasa train for the tutorial example. 2024-04-09 15:06:51 INFO rasa.tracing.config - No endpoint for tracing type available in endpoints.yml,tracing will not be configured. 2024-04-09 15:06:52 INFO rasa.cli.train - [info ] Started validating domain and training.