 |
 |
|---|
Dieser Prototyp erstellt einen ersten Vorschlag für eine Version eines Textes in Leichter Sprache, der zum Beispiel an eine Übersetzerin oder einen Übersetzer für Leichte Sprache weitergegeben werden kann. Der generierte Text ist oft noch nicht perfekt. Wir empfehlen eine nachträgliche Prüfung auf inhaltliche Korrektheit und Verständlichkeit. Insbesondere ersetzt dieser Prototyp nicht die Prüfung durch Personen, die der Zielgruppe Leichter Sprache angehören.
This prototype creates an initial proposal for a version of a text in simple language that can be passed to a translator for simple language. The generated text is often not yet perfect. We recommend a subsequent check for correctness of content and comprehensibility. In particular, this prototype does not replace checking by people who belong to the Simple Language target group.
Locally-run tool to simplify German into "Leichte Sprache" (easy language) based on Large Language Models (LLMs).
The browser interface is designed for intuitive use, allowing experimentation with different models and approaches. The implementation focuses on simplicity, low-level components and modularity, allowing developers and Python enthusiasts to modify and build upon.
Key Features
- Simplifies complex German text into easy-to-understand language
- Browser-based interface for intuitive usage
- Local execution for privacy and control
- Configurable parameters for output customization
Download or clone the repository.
Follow the installation instructions
Two different frontends are available in this expanded version of the repository:
- app.py – A simplified frontend for use by users without prior technical knowledge.
- app-Front-technisch.py – An advanced frontend for technical developers with access to advanced settings and test scenarios.
Both frontends use the same backend logic but differ in structure and presentation. Switching between the two versions is done by calling the correct file in the terminal with the following command
$ python3 app.pyor (for the advanced frontend)
$ python3 app-Front-technisch.py- Start the tool [*]
$ python3 [name-of-file].py- Open the provided URL on your web browser
- Write or paste a text to simplify
- adjust the contrast or font size if necessary
- Enjoy
- Model: Select the desired Model
- Use Rules (checkbox): Click for adding Leichte Sprache rules to the prompt sent to the LLM.
-
Top k: Ranks the output tokens in descending order of probability, selects the first k tokens to create a new distribution, and it samples the output from it. Higher values result in more diverse answers, and lower values will produce more conservative answers.
-
Temp: This affects the “randomness” of the answers by scaling the probability distribution of the output elements. Increasing the temperature will make the model answer more creatively.
[*] If you chose the installation with a virtual environment, remember to activate it before starting the application by running $ source .myvenv/bin/activate
Performance consideration: On notebooks/PCs with dedicated GPUs, all the set models should run properly and fast. On a standard notebook, or if you encounter any memory or performance issues, prioritize the models based on llama 3.2 as they are smaller than the models based on llama 3.1 and need less hardware requirements.
There are additional command line input tools that you can use to batch process data.
Before committing, format the code using Black:
$ black -t py311 -S -l 99 .Linters:
- Pylance
- flake8 (args: --max-line-length=100 --extend-ignore=E401,E501,E741)
For more detailed logging, set the LOG_LEVEL environment variable:
$ export LOG_LEVEL='DEBUG'
This is an artefact from a collaboration between DigitalAgentur Brandenburg and KI-Servicezentrum Berlin-Brandenburg. DigitalAgentur Brandenburg is funded by the Ministry for Economic Affairs, Labour and Energy of the State of Brandenburg. KI-Servicezentrum Berlin-Brandenburg is funded by the Federal Ministry of Education and Research under the funding code 01IS22092.