400: Unsupported value: 'temperature' does not support 0 with this model

[stepanov1975]

Hi. When trying to use o4-mini model getting error:
paperless-gpt | time="2025-05-31T09:23:07Z" level=error msg="Error processing document 2405: error getting response from LLM: all retry attempts failed, last error: API returned unexpected status code: 400: Unsupported value: 'temperature' does not support 0 with this model. Only the default (1) value is supported." document_id=2405

[thiswillbeyourgithub]

That's weird I don't see "temperature" being mentioned anywhere in that repo so it does not appear to be set.

[benjaminbalo]

i got the same error with gpt-5-nano

time="2025-08-10T06:44:37Z" level=error msg="Error processing documents: Document 800: error getting response from LLM: all retry attempts failed, last error: API returned unexpected status code: 400: Unsupported value: 'temperature' does not support 0 with this model. Only the default (1) value is supported."

[inboxcda]

Getting something similar from docling logs. Integrated docling with paperless-get.

'The following generation flags are not valid and may be ignored: ['temperature']. Set TRANSFORMERS_VERBOSITY=info for more details.'

[R4mboHKM]

Having the same problem with gpt-5-mini

Settings via environment:
LLM_MODEL: gpt-5-mini
LLM_PROVIDER: openai
OPENAI_API_KEY: "..."

Error Message:
time="2025-09-16T11:37:46Z" level=error msg="Error processing documents: Document 2048: error getting response from LLM: all retry attempts failed, last error: API returned unexpected status code: 400: Unsupported value: 'temperature' does not support 0 with this model. Only the default (1) value is supported."

Manual OCR via Mistral works.
gpt-4o works without error, so its model-specific (in my case)

[stepanov1975]

Hi. Is there any progress. Still failing with same error for gpt-5-mini

[R4mboHKM]

If it helps, as a workaround I switched to OpenRouter.
Using these env Variables:

LLM_LANGUAGE: German (Choose your language)
LLM_MODEL: openai/gpt-5-mini
LLM_PROVIDER: openai
OPENAI_BASE_URL: https://openrouter.ai/api/v1
OPENAI_API_KEY: *** (OpenRouter API Key)

[Malli88]

If it helps, as a workaround I switched to OpenRouter. Using these env Variables:

LLM_LANGUAGE: German (Choose your language) LLM_MODEL: openai/gpt-5-mini LLM_PROVIDER: openai OPENAI_BASE_URL: https://openrouter.ai/api/v1 OPENAI_API_KEY: *** (OpenRouter API Key)

Good solution. Works for me, but the main problem should be fixed anyway.

[adfaklsdjf]

#573 includes a bunch of features, one of them being a new VISION_LLM_TEMPERATURE variable for setting the temperature value for the vision model. Setting this to 1.0 has gpt-5-mini working for OCR for me.

The image I built is at adfaklsdjf/paperless-gpt:pr-573 on dockerhub.