v2.1.9

[clusterzx]

Full Changelog: v2.1.8...v2.1.9

Added a new Settings option:
Brought back num_ctx (100000) for Ollama.

"Use existing Correspondents and Tags".

If you choose yes, all your tags and correspondents will be added to the prompt for later data matching and providing better consistent results. BUT it will increase your token length and will be costlier and for Ollama much more possible to run into issues.
When you have many preexisting data it will be somtimes the case that the prompt only consists of your data and nothing more. So pay attention!

---

This discussion was created from the release v2.1.9.

[rjbutler]

You may also need to explicitly filter out tags that don't exist from the response since LLMs aren't all that great at following that type of instruction.

[mamema]

frankly i haven't expected that feeding all tags and correspondents into ollama will work. Feels like a "shut up" tasks, and it is:

[ERROR] fetching document 26144: connect ECONNREFUSED 192.168.0.30:8873 2025-01-13T20:13:33: PM2 log: Stopping app:paperless-ai id:0 [DEBUG] Received SIGINT signal. Starting graceful shutdown... [DEBUG] Closing database... [DEBUG] Database closed successfully [DEBUG] Database closed successfully 2025-01-13T20:13:33: PM2 log: App name:paperless-ai id:0 disconnected 2025-01-13T20:13:33: PM2 log: App [paperless-ai:0] exited with code [0] via signal [SIGINT] 2025-01-13T20:13:33: PM2 log: pid=44 msg=process killed 2025-01-13T20:13:33: PM2 log: PM2 successfully stopped

from 14 documents one was tagged and processed, within the last two hours,. but this in a very good quality, but the above seems to have stopped the whole process

[clusterzx]

Looks like the tags and correspondents are too much to handle if you have the option on.

Ahh missed the first sentence.
Yep too much to handle.

[joestump]

I think I’ve been having a similar issue. Switched to OpenAI and it works fine.

[mamema]

....and i have this "enter username" situation again, without reseting/deleting anything

[clusterzx]

no, thats the point! I don´t want to change anything and i din´t start the setup process (again). What i´m trying to say is, i have the interface open and "out of the blue" suddenly i´m confronted with the reenter password page again

Then maybe your persistent volume did not match to the new installation, or maybe you dont set it as persistent volume. The setup process only pops out if 1. the config (.env) misses some inforamtion OR the paperless-ngx instance was not running before paperless-ai was started.

[mamema]

i've using bind mount. But that paperless-ngx isn't running sometimes could be the case. I'm looking at it again.

[clusterzx]

i've using bind mount. But that paperless-ngx isn't running sometimes could be the case. I'm looking at it again.

Friend of mine had exact the same issue and we were wondering for almost an hour why. Turned out paperless-ai starts quicker and before paperless-ngx

[mamema]

that paperless-ai isn't working if ngx isn't running, is a nobrainer, but reentering password isn't obvious.

[mamema]

got this also during save of settings (which takes sometimes ages with ollama and i don't know why) and then rebooting ollama

[mamema]

i have this with ollama (too slow...)
[ERROR] processing document 26140: Error: [ERROR] Document analysis failed: timeout of 1200000ms exceeded at processDocument (/app/server.js:97:11) at runNextTicks (node:internal/process/task_queues:65:5) at listOnTimeout (node:internal/timers:575:9) at process.processTimers (node:internal/timers:549:7) at async scanInitial (/app/server.js:170:24) at async startScanning (/app/server.js:279:5)

is there something i can do on ollama and/or paperless-ai side?

[mamema]

i just discovered, that on every "save settings" task this block gets added:

`Return the result EXCLUSIVELY as a JSON object. The Tags and Title MUST be in the language that is used in the document.:

" {
"title": "xxxxx",
"correspondent": "xxxxxxxx",
"tags": ["Tag1", "Tag2", "Tag3", "Tag4"],
"document_date": "YYYY-MM-DD",
"language": "en/de/es/..."
}`

so there are duplicates of this block, when saved multiple times. Also if someone (like myself) changed the date format via prompt instruction to DD-MM-YYYY this seems a problem then.....