Ollama KNIME - using LLMs to split names and extract them into CSV - KNIME Forum (84090)

<p>Use LLM (Ollama/Llama3.2) to process text data in Chunks and collect a structured output on your local machine</p><ul><li><p>you need Ollama installed and ready to run on your machine, also you should have GPt4All installed</p></li><li><p>tested it with Llama 3.2</p></li><li><p>The text you want to clear / be processed will be stored in chunks in JSOn structures to the LLM will understand the content and you can refer to the columns in the Prompt</p></li><li><p>There will be 10 lines at a time so as not to overflow the memory of the model</p></li><li><p>The Question will consist of the Prompt and then the JSOn file itself</p></li><li><p>The results will either be a CSV structure or sometimes a CSV structure enclosed in backticks - so there is a fork in the workflow to account for both. You might have to adapt this for your own data</p></li><li><p>The columns with the extracted CSV will then be written to actual CSVs and imported back</p></li><li><p>Key is that you tell the LLM to preserve the original row_key so as to be able to join it together</p></li><li><p>obviously some more work on the prompt or the data preprocessing might be necessary - you could also try and employ a (well) LLM to do that</p></li><li><p>An Apple Silicon machine (M1) works reasonably fast in order to let this run locally</p></li></ul>