Running DeepSeek Locally: Capabilities and Challenges

DeepSeek ist qualitativ hochwertig, aber leider zensiert und datenschutztechnisch fragwürdig. c't 3003 versucht, das vollständige Modell lokal laufen zu lassen.

The text discusses the capabilities and challenges of running the AI language model DeepSeek locally, as opposed to using its online version, which is often criticized for censorship and data privacy concerns. The author, Jan-Keno Janssen, explores the feasibility of achieving similar performance on personal hardware, highlighting that while DeepSeek can function locally, it requires high-end GPUs or significant CPU resources. The local version of DeepSeek reportedly does not exhibit the same censorship issues as its cloud counterpart and can handle complex tasks effectively. However, running such large models locally can be costly and technically demanding.

DeepSeek is known for its quality but is criticized for censorship and privacy issues.
Local installation requires powerful hardware, such as multiple high-end GPUs or CPUs with large RAM.
The local version has less censorship and performs comparably to the online version.
Running large models locally is expensive and complex, with a need for specialized setups.

What are the limitations of running DeepSeek locally?

Running DeepSeek locally requires significant hardware resources, such as multiple high-end GPUs or a powerful CPU with large amounts of RAM. This can be costly and technically complex.

How does the local DeepSeek version compare to the online version?

The local version of DeepSeek reportedly has less censorship and can produce similar or better results compared to the online version, particularly in handling complex queries.

Is it safe to run DeepSeek locally given its origin?

While there are concerns about the model's origin, experts suggest that the risk is low since DeepSeek operates as data rather than executable code, making it less likely to pose security threats.