"Prompt OpenAI LLMs" example issues

[roberting]

I am playing with the “Simple OpenAI Prompt” Workflow provided here: 1.1 - Simple OpenAI Prompt – KNIME Community Hub.
I downloaded the Workflow and entered my API key, leaving everything else default.
Of the 11 rows of prompts only 4 were answered (the text summarization and Text completion). All other were left blank. Am I missing something here ?

[mlauber71]

@roberting it is difficult to say something about this. Were these the first questions, are there any error messages on the log? Can you check your account if you have sufficient credits?

In the meantime might I interest you in an approach to use local LLMs

KNIME, AI Extension and local Large Language Models (LLM)

Creating a Local LLM Vector Store from PDFs with KNIME and GPT4All

[roberting]

Hey @mlauber71 , thanks for the quick reply and the links.
while I have played around with ChatGPT, I am new to the (technical) background of LLM and generative AI.
I have paid 20$ into the account just to check this out. The outcome is like in the screenshot; the empty fields seem to be Task related.

[roberto_cadili]

Hi @roberting,

I’ve run the same workflow you downloaded from the KNIME Community Hub without any changes. Good news: I see the response .

I’m sure you have the response too, you just need to see where it is (check my GIF).

Why does it happen? Because in some cases the response of the LLM model is introducing a line break before providing the actual answer. The LLM Prompter node does not operate any transformation on the output provided by the LLM. So what you get is the raw output of the LLM.

From my experience, a line break is more likely to be included if you’re asking to solve a task for which the answer can be more easily presented as a sort of list or bullet points. In the example workflow, this happens indeed where we’re asking to perform a classification task (multi-label output, and single label output).

As for the billing part, 20 $ is definitely an amount that you will take a while to exhaust if you’re using ChatGPT for text generation and similar tasks . So far, I’ve personally spent less than 2 dollars (I’m a relatively frequent user).

Mind, however, that if you’re using the OpenAI DALL-E View node for image generation, you are likely to exhaust your credit sooner. OpenAI charges much more for DALL-E. Check the pricing here.

Hope it helps,

Happy KNIMEIng,
Roberto

[roberting]

hi @roberto_cadili
thanks for your replay - sometimes it’s the minute details that cause the biggest headaches.