Trouble getting auto tagging to work

[Laephis]

Hi,
Apologies up front if I'm missing something obvious, but I can't get the auto tagging feature to work. Manual tagging via 'paperless-gpt' works fine - it shows up on the paperless-gpt dashboard and I'm able to generate tags and scan ocr. If I tag something with 'paperless-gpt-ocr-auto' or 'paperless-gpt-auto', nothing happens, the document just remains unchanged. Here's the environment options I'm using if that helps:

paperless-gpt:
image: icereed/paperless-gpt:latest
environment:
PAPERLESS_BASE_URL: 'http://192.168.1.184:8002'
PAPERLESS_API_TOKEN: 'token placed here'
PAPERLESS_PUBLIC_URL: 'http://paperless.mydomain.com' # Optional
MANUAL_TAG: 'paperless-gpt' # Optional, default: paperless-gpt
AUTO_TAG: 'paperless-gpt-auto' # Optional, default: paperless-gpt-auto
LLM_PROVIDER: 'ollama' # or 'ollama'
LLM_MODEL: 'llama3.2' # or 'llama2'
# OPENAI_API_KEY: 'your_openai_api_key'
# Optional - OPENAI_BASE_URL: 'https://litellm.yourinstallationof.it.com/v1'
LLM_LANGUAGE: 'English' # Optional, default: English
OLLAMA_HOST: 'http://192.168.1.108:11434' # If using Ollama
VISION_LLM_PROVIDER: 'ollama' # (for OCR) - openai or ollama
VISION_LLM_MODEL: 'llava' # (for OCR) - minicpm-v (ollama example), gpt-4o (for openai), etc.
AUTO_OCR_TAG: 'paperless-gpt-ocr-auto' # Optional, default: paperless-gpt-ocr-auto
LOG_LEVEL: 'info' # Optional: debug, warn, error
OCR_LIMIT_PAGES: '0'

[icereed]

Hi, can you send some logs?

thank you 🙏

[Laephis]

Sure thing - what logs would be most helpful?

I added the auto tags to a new document and looked at the paperless.log output but only saw this:

[2025-01-11 08:05:00,308] [DEBUG] [paperless.classifier] Gathering data from database...

[2025-01-11 08:05:00,418] [DEBUG] [paperless.classifier] 66 documents, 14 tag(s), 0 correspondent(s), 18 document type(s). 0 storage path(es)

[2025-01-11 08:05:00,418] [DEBUG] [paperless.classifier] Vectorizing data...

[2025-01-11 08:05:01,828] [DEBUG] [paperless.classifier] Training tags classifier...

[2025-01-11 08:05:06,827] [DEBUG] [paperless.classifier] There are no correspondents. Not training correspondent classifier.

[2025-01-11 08:05:06,827] [DEBUG] [paperless.classifier] Training document type classifier...

[2025-01-11 08:05:09,829] [DEBUG] [paperless.classifier] There are no storage paths. Not training storage path classifier.

[2025-01-11 08:05:09,832] [INFO] [paperless.tasks] Saving updated classifier model to /usr/src/paperless/data/classification_model.pickle...

then looked at the paperless-gpt logs in docker and didn't see anything stand out:

[GIN] 2025/01/11 - 14:08:20 | 200 | 127.377827ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:21 | 200 | 113.493114ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:22 | 200 | 147.87771ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:23 | 200 | 122.022375ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:24 | 200 | 161.650681ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:26 | 200 | 159.943755ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:27 | 200 | 125.001468ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:28 | 200 | 136.960532ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:29 | 200 | 169.399435ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:30 | 200 | 146.455083ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:31 | 200 | 163.332357ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:33 | 200 | 131.810135ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:34 | 200 | 152.239092ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:35 | 200 | 128.135978ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:36 | 200 | 143.237208ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:37 | 200 | 148.774647ms | 192.168.1.116 | GET "/api/documents"
[GIN] 2025/01/11 - 14:08:38 | 200 | 138.157829ms | 192.168.1.116 | GET "/api/documents"

[icereed]

Stupid question of me: did you try to restart paperless-gpt before? Just one hypothesis that I have is that their tags were not created before paperless-gpt was started and hence the ID could not be read. 🫠

[Laephis]

I restarted the docker container (separate from the main paperless-ngx docker compose setup) and things are showing up in the log files now. Thanks!
It's not creating any tags, though, and the metadata that's being generated is 100% different from anything in the document - weirdly different.

Using paperless-gpt as manual tag
Using paperless-gpt-auto as auto tag
Using paperless-gpt-ocr as manual OCR tag
Using paperless-gpt-ocr-auto as auto OCR tag
[GIN-debug] [WARNING] Creating an Engine instance with the Logger and Recovery middleware already attached.

[GIN-debug] [WARNING] Running in "debug" mode. Switch to "release" mode in production.

• using env: export GIN_MODE=release
• using code: gin.SetMode(gin.ReleaseMode)

[GIN-debug] [WARNING] You trusted all proxies, this is NOT safe. We recommend you to set a value.
Please check https://pkg.go.dev/github.com/gin-gonic/gin#readme-don-t-trust-all-proxies for details.

time="2025-01-11T20:40:48Z" level=warning msg="No valid title found for document 2, skipping."
...
time="2025-01-11T20:41:18Z" level=warning msg="No valid title found for document 42, skipping."
...
time="2025-01-11T20:44:32Z" level=warning msg="No valid title found for document 52, skipping."

[icereed]

Which model are you using? Sometimes too small models are easily confused.

[Laephis]

Llama3.2 which is only 3b, so yeah, on the smaller side. I'll try Llama3.1 (8b) and see if that helps. (Hopefully able to run it OK.)

Thanks for your help!

[icereed]

You could also try this one:
https://ollama.com/library/phi4

It’s quite smart yet small.