deepseek-ai

/

DeepSeek-V3-Base

like 1.02k

Follow

DeepSeek 4.5k

Preferably Apache 2.0 :)

Preferably MIT :)

Preferably public domain :)

Seems like the license's there:

https://huggingface.co/deepseek-ai/DeepSeek-V3-Base/blob/main/LICENSE-MODEL

But since they uploaded both the code and the weights to the same repository, their team has to contact HuggingFace to add their custom license to HF's list of available licenses.

The license seems to be rather limited, especially for researching safety, creating roleplay-focused models, or improving the capabilities of existing models.

• You must enforce restrictions listed in Attachment A, such as law violations, topics that may be considered harmful to the minors by the international court, "discriminating against or harming individuals", a vague definition of "inappropriate content", etc. This could be borderline impossible for some clients without a dedicated legal team and/or the means to deploy mandatory filtering endpoints.
  «The restrictions set forth in Attachment A are considered Use-based restrictions», «Attachment A», «Use Restrictions»
• You can't use the model to generate synthetic datasets for training or finetuning purposes to create properly permissive models under permissive licenses.
  «"Derivatives of the Model" means... output of the Model, to the other model, in order to cause the other model to perform similarly to the Model, including... methods based on the generation of synthetic data by the Model for training the other model.»
  «Even though downstream derivative versions of the model could be released under different licensing terms, the latter will always have to include - at minimum - the same use-based restrictions as the ones in the original license (this license).»

However, they don't claim ownership over the outputs, allowing you to use them for, say, novel writing or blog posting without any legal consequences, which is nice.
«The Output You Generate. Except as set forth herein, DeepSeek claims no rights in the Output You generate using the Model. You are accountable for the Output you generate and its subsequent uses. No use of the output can contravene any provision as stated in the License.»
It seems the output itself isn't affected by the "Use Restrictions", as they only regulate the usage of the weights, not the final output (except when the resulting work is considered a derivative), but I can't say for sure, a legal expert should probably correct me here.

In general, unfortunately, it seems like the model was developed for business customers and cannot be touched by enthusiasts of our community without breaking the license terms.