Skip to content

Commit

Permalink
Merge pull request #12246 from nextcloud/backport/12245/stable30
Browse files Browse the repository at this point in the history
[stable30] More additions about AI
  • Loading branch information
marcelklehr authored Sep 27, 2024
2 parents be26fd9 + 5513eaf commit 717b490
Show file tree
Hide file tree
Showing 2 changed files with 17 additions and 1 deletion.
3 changes: 2 additions & 1 deletion admin_manual/ai/app_assistant.rst
Original file line number Diff line number Diff line change
Expand Up @@ -38,11 +38,12 @@ Artificial intelligence at Nextcloud is built in a modular way, allowing you to

**Note**: At Nextcloud we focus on creating on-premise AI apps that run fully self-hosted on your own servers in order to preserve your privacy and data sovereignty. However, you can also offload these resource-heavy tasks to an :ref:`"AI as a Service" provider<ai-ai_as_a_service>`.

**Note**: When using our on-premise AI apps, make sure you have a GPU with enough VRAM that fits all the features you need. For each app documented here you will find its hardware requirements.

Machine translation
~~~~~~~~~~~~~~~~~~~

In order to make use of machine translation features in the assistant, you will need an app that provides a translation backend:

* :ref:`translate2 (ExApp)<ai-app-translate2>` - Runs open source AI translation models locally on your own server hardware (Customer support available upon request)
* *integration_deepl* - Integrates with the deepl API to provide translation functionality from Deepl.com servers (Only community supported)

Expand Down
15 changes: 15 additions & 0 deletions developer_manual/digging_deeper/task_processing.rst
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,21 @@ The following built-in task types are available:
* Output shape:
* ``output``: ``ListOfImages``

LLM Prompts and multilingual I/O
################################

When writing prompts for the TextToText task type in your apps, we recommend testing it with at least

* OpenAI GPT-3.5
* Llama 3.1

Also, make sure that you instruct the model to use the correct language in its output. By default most models will answer in English if the main prompt is in English, even though the source data is in another language.
A tweak to make sure of this is to instruct the model as follows:

.. code-block:: php
"Detect the language used in the text and make sure to answer in the same language without mentioning the language explicitly."
Input and output shapes
~~~~~~~~~~~~~~~~~~~~~~~

Expand Down

0 comments on commit 717b490

Please sign in to comment.