How to

Please stick to the following guidelines when uploading resources.

How to upload models

• Create a model repo using the following naming schema: <use-case>-<pre-trained-model-name>-<training-type>-<version>
  • E.g., du-qwen3-4b-sft-v0
  • Keep the <pre-trained-model-name> as simple as possible (e.g., Apertus-8B-Instruct-2509 can become Apertus-8b)
  • Mention the complete pre-trained-model-name in the model card
  • Important: If this model will be deployed and used in production please append the "latest" keyword to the name and remove it from the repo that is currently marked as "latest"
• Upload the files to the created repo
• Write a model card with at least the following information
  • Use Case (e.g., DU, NLG, ...)
  • Pre-Trained Model Name
  • Training Type (e.g., SFT, DPO, ...)
  • Used Hyperparameters
  • Version
  • Used dataset(s)
  • Where was it trained (e.g., Fireworks AI, Google Colab, ...)
    • If trained on Colab please add the Link to the Notebook (both to Colab and GitLab)
• Add the repo to an existing collection, or create a new one if none applies

How to upload datasets

• Create a dataset repo using the following naming schema: <use-case>-<training-type>-<version>
• Upload the files to the created repo
• Write a dataset card with at least the following information
  • Summary
  • Supported tasks
  • Dataset structure
  • Dataset creation
• Example for a dataset card: https://huggingface.co/datasets/novomind/nlg-sft-v1