Text Lab is a secure, interactive suite of advanced Artificial Intelligence and Natural Language Processing (NLP) tools designed to run directly on the University of Bern's High-Performance Computing cluster (UBELIX).
For comprehensive instructions, tutorials, capabilities, and best practices, please visit the official User Guide: https://text-lab.dsl.unibe.ch/
The application features a modular, secure architecture offering the following tools:
- ** Audio Transcription:** Highly accurate multi-lingual transcription using Whisper. Includes Speaker Diarization, VAD pre-filtering, and dedicated support for Swiss German.
- ** AI Chat:** A private AI assistant (powered by Ollama) that allows you to upload and interact directly with your own documents (PDFs, text, datasets).
- ** Advanced OCR:** Extract text, complex tables, and layouts from images and PDFs using a choice of state-of-the-art vision models (GLM-OCR, olmOCR, PaddleOCR, EasyOCR).
- ** AI Data Visualiser:** Upload your tabular data and instruct an AI agent (via an internal MCP server) to automatically write Python code and generate interactive plots.
- ** Knowledge Graph Generator:** Process collections of scientific papers using Grobid to automatically extract metadata, citations, and research topics via LLMs, visualizing them as an interactive network graph.
- ** Topic Modeling:** Extract topics from raw text data using SOTA models.
Because all models run on university hardware via the Open OnDemand platform, your sensitive research data never leaves the university network. The application utilizes zero-footprint, ephemeral session states to ensure maximum security on shared infrastructure.
The app is accessible via UBELIX Open OnDemand and is available only within the Unibe internal network (or via VPN).
When launching the app from the Data Science Lab Services menu, you can specify several parameters to tailor its execution:
- Job Time (hours): Specify the maximum duration for the app to be active. It will automatically terminate after this time to ensure fair resource allocation.
- GPU Type: Request a specific GPU (e.g., RTX 4090, A100).
- SLURM Partition: Select the
gpupartition. - Quality of Service (QoS): Use
job_gpu_preemptablefor quick tasks. Note that preemptable resources can be reclaimed by the system if needed. Usejob_gpuif you have a specific allocation. - Number of GPUs: Typically
1is sufficient, but more can be requested if running very large LLMs in the Chat tool.
For more detailed guidelines on job submission, please refer to the HPC Documentation.
src/pages/: Streamlit UI frontend interface files.src/core/: Backend business logic, LLM interaction, MCP server, and data processing engines.src/assets/: Application icons and logos.docs/: MkDocs documentation site source code.template/&form.yml: Open OnDemand deployment configurations.
Maintained by the Data Science Lab (DSL) at the University of Bern.
For support with the app, bug reports, or related NLP services, please contact: support.dsl@unibe.ch
If you'd like to be informed about updates and changes to Text Lab, please subscribe to the following Mailing List