Llama-2-13B · 13B · 2023 · Llama 2 Community LicenseLeoLM
The first open suite of German-speaking foundation models — by LAION and hessian.AI, trained on the 42 supercomputer.
open chat →the ai ecosystem of hesse · the registerlive·the pulse →
open models
demotalk to the
open models.
Two open language models trained on hessian.AI's 42 supercomputer — LeoLM (2023) and Occiglot (2024). Both run here on hessian.AI's inference infrastructure at TU Darmstadt. A light demo: a short chat, no account, nothing stored.
Mistral-7B-v0.1 · 7B · 2024 · Apache-2.0Occiglot
A European multilingual model for the five largest EU languages — by the Occiglot research collective, trained with hessian.AI compute.
- model card · hugging face ↗
- EU5 7B collection ↗
- announcement · occiglot.eu, mar 2024 ↗
- technical report v0.1 ↗
- github ↗
Research models. Both are older, small research models (7B/13B) without comprehensive safety tuning. Answers can be wrong, outdated or inappropriate — don't rely on them unverified, and please don't enter personal data. Messages are processed transiently on hessian.AI infrastructure at TU Darmstadt and are not stored by us. Privacy details on /about.