Apertus 70B: Truly Open - Swiss LLM by ETH, EPFL and CSCS
8 days ago
- #AI
- #Open Source
- #Language Model
- Apertus LLM is a fully-open multilingual language model developed by the Swiss National AI Institute (SNAI).
- Users must agree to indemnify ETH Zurich and EPFL against third-party claims arising from the use of Apertus LLM.
- The model processes Personal Data, and users are advised to apply an output filter every six months to remove such data.
- Apertus supports over 1000 languages, uses open weights and data, and respects opt-out consent of data owners.
- The model is a decoder-only transformer pretrained on 15T tokens with a curriculum of web, code, and math data.
- Apertus can be used via transformers or vLLM, supports long context up to 65,536 tokens, and has agentic usage capabilities.
- Evaluation results and training details, including architecture and hardware, are openly available.
- Limitations include potential factual inaccuracies, biases, and the need for critical evaluation of generated content.
- Legal aspects cover EU AI Act compliance, data protection, and copyright requests.
- Contact and citation details are provided for further inquiries and academic references.