[Question]: Parsing process stuck at keyword, question generation and running RAPTOR

[predoctech]

Describe your problem

Under "General" template there are parameters for generation of keyword, question, and running RAPTOR. However in demo site when trying to instantiate any of those process the parsing will get stuck. So with all those parameters set to 0/off the parsing process can complete, otherwise the process will hang when any of those steps are initiated. In the case of RAPTOR there has been error messages saying ".... LLM is needed to run RAPTOR" but both Chat and Embedding models have already been set up inside Settings.
So what are the requirements in order to run those steps during a dataset parsing? Both the default Deekseek models and Gemini models have been used for Model settings but both resulted in the same hung process.
Anyone with success in running these steps in Dataset parsing please share your experience. Thanks.

[KevinHuSh]

An LLM API key is required for those processes on demo. The default free Deepseek model is not feasible.

[predoctech]

Thanks Kevin.
What kind of LLM model is needed? And should that be the one defined under the Chat model settings? FYI besides Deepseek I've also tried with Gemini models (with API key obviously) without any success and same findings. Please further advise.

[KevinHuSh]

I've checked on demo site. It takes long time but eventually success.
Try Gemini again?

[predoctech]

Following your suggestion I restarted a task on a 16 chunk dataset and now the RAPTOR process is still onging after 6 hours. Progress is shown as 50%, is that the normal/expected "long time"? Would a local deployment improve that in any way because honestly at that pace this function is non-usable. Thanks.

[coldrainlin]

I encountered the same issue. When working with a 7MB JSON file and calling the paid APIs for QwenEmbedding and DeepSeekV3Chat while simultaneously attempting to use RAPTOR to enhance retrieval or a custom delimiter, there is an extremely long time with no progress. The speed is incomparable to when RAPTOR and the custom delimiter are not used. This makes it very difficult to use.
If you find a relevant solution, please be sure to contact me. Thank you.