[Question]: After using the recall - enhanced RAPTOR strategy, how should the <think></think> tags generated in the parsed chunks be removed?

[12moona]

Describe your problem

After using the RAPTOR to enhance retrieval how should the tags generated in the parsed chunks be removed?

[KevinHuSh]

It's already been fixed. Pull the nightly version of docker image again.

[12moona]

It's already been fixed. Pull the version of docker image again.nightly

@KevinHuSh Hello! I have already pulled the latest version of the image: nightly, but there are still tags during the test. May I ask what the reason is?

[KevinHuSh]

You needed to parse it again, didn't you? And it has cache of LLM response for 24H.

[12moona]

@KevinHuSh Yes, I re - parsed the files. Moreover, I also uploaded other different files for parsing, but the tags still appear.

[12moona]

I've checked the relevant backend code. It seems that there haven't been any recent changes, and it's no different from the code in version v0.16.0.

[KevinHuSh]

So, you did not pull the image again, did you?

[12moona]

@KevinHuSh ,Yes, the latest image I pulled, which is the "nightly" version, was from four days ago. However, the line of code was added two days ago, so it's not included in the "nightly" version. I've already added it now, but the problem hasn't been completely resolved yet, and there are still tags.