hey @dryn07,
Sorry that you’re experiencing that! A few questions so I can try to replicate on my end.
what embedding model are you using?
I would assume you’re using the WordPress database?
what type of import are you trying to edit (wordpress import, sitemap import, manual entry)?
thanks,
maxwell
Thread Starter
dryn07
(@dryn07)
Man, it removed my whole knowledge base…. what the hell… I just wanted to edit one entry
hi @dryn07,
I am not able to replicate this. All edits of all types work perfectly fine for me. To be honest, its not even possible for it to delete all entries from clicking the edit button. The edit and delete all functions are not connected in any way.
Do you have the answers to my questions above?
Thanks,
Maxwell
apologies again youre experiencing this. Im actively investigating and doing anything I can to replicate. Let me know any further info you may have.
hey @dryn07,
I believe I may be getting close to uncovering the bug. Were all of your entries large manually submitted entries?
Thread Starter
dryn07
(@dryn07)
I tried to import PDF, the I got:
ERROR
Failed to generate embedding for content
Processing returned false – item may be empty or invalid
Thread Starter
dryn07
(@dryn07)
Sorry I did not see your replies 😀
I tried ada and google embedding.
Yes, it is wp database becasue it is not a big data (I think). My biggest entry is a 100 row FAQ. But it happens with a one row entry as well.
I used a plain text import, now I tried a pdf :/
apologies, I dont want to get confused because now it seems youre asking about something else.
I was asking about your original issue. What type of content you had that you tried to edit which deleted everything?
it also now appears youre trying to resubmit the information via PDF but experiencing issue resubmitting. Are you able to link the PDF here? It must be an OCR PDF to work
Thread Starter
dryn07
(@dryn07)
Sorry, I wrote a bit scattered. So:
I’ve been using the plugin for about 1-2 years, and until now I’ve been uploading “Direct Content” data, because the knowledge was precisely defined. There were about 20 entries, but mostly smaller pieces of data — it was just easier to maintain this way. There were no issues until now (although I did have some minor editing problems with the older mxchat version as well), and I’ve just logged into the redesigned version for the first time. I clicked the pencil icon on one of the entries, changed a couple of minor things, and upon saving it immediately threw an error. I reloaded the page and my entry no longer existed. I tried with another one — same error, and after reloading it was completely empty. I had turned on debug mode beforehand, but I don’t see anything unusual in it.
I’ve now compiled all the material into a single md file, which is only 260 lines (16k chars), so I wouldn’t call it a huge amount of data. From this I created a clean PDF and tried to import it. I ran an accessibility check on the PDF, OCR’d it, so in theory everything is fine. But the error still appears.
I’ve adjusted the embedding, LLM, etc., but the error persisted.
I’m using LightSpeed cache, if that matters.
PDF:
https://drive.google.com/open?id=1l54BceDHN_gOlHZ0UrD6FAXEFnQi5Gst&usp=drive_fs
I could replicate it from a backup, here is a VIDEO:
https://drive.google.com/file/d/1wjU9ANCLDKrNPvIqhmL_rGV5AP50RcF_/view?usp=sharing
hey @dryn07,
This was extremely helpful. I apprecaite you taking the time to share and apologies again for the inconvienence. What occured was that older versions of mxchat would a source URL with manual entries such as mxchat://manual – it was somewhat of a bug. The problem is when you tried to edit the new pipeline does not support a mxchat URL type so it failed. The error compounded because you can’t edit vectors directly, it’s not possible, so what actually occurs when you edit is that it’s deleting the original vector and then revectoring a new one with the updated content – since all manual entries shared the same source URL it deleted all entries associated with that source URL and then failed to revector because of the unsupported mxchat:// URL type. I’m updating it now to accept those old URL types so this doesn’t happen to anyone in the future.
As for resubmitting the new PDF. I’m not sure if you are trying to submit the google drive URL directly. That will not work because the PDF import option must be a direct source link to an actual PDF file (ending in .pdf). The google drive link is actually an HTML page with a PDF viewer and a download link. In the next release I will have it support google drive PDF links like this. For now the option would be to upload the PDF to like your media library and use that link or use the direct upload PDF option and try those.
Thanks!
Hey @dryn07,
Version 3.1.7 was just released. It includes the ability to use your google link, will show better errors when something goes wrong, and fixes the edit issue on older manual entries with the mxchat url.
Thanks,
Maxwell
Thread Starter
dryn07
(@dryn07)
Thank you for your detailed reply!
The mentioned pdf is a simple .pdf file. Should I make a video? 😀
Wern I get to my pc I will try the update as well.
