README.md

metadata

language:
  - en
pretty_name: community base_model sprint data
tags:
  - metadata

Base Model Metadata Sprint

Description

Join us in improving the discoverability and understanding of models on the Hugging Face Hub by adding base_model metadata! This sprint aims to enhance the information available for models derived from, fine-tuned on, or quantized versions of existing base models.

🤗 Strong contributions will win prizes!! 🤗

Why It Matters

Adding base_model metadata helps users:

1. Easily find models derived from specific architectures
2. Understand the lineage and potential capabilities of a model
3. Make informed decisions when choosing models for their projects
4. Enables a useful Model Tree on the model repo!

How to Contribute

1. Access the Sprint CSV
   • Visit the Sprint CSV file on the Hugging Face Hub here
   • This CSV contains models that need to be checked for base_model metadata

2. Choose a Model

   • Select a model from the CSV that hasn't been processed yet (PR_CREATED is false)
   • Prioritize models with higher 'likes' counts for maximum impact

3. Check Existing PRs

   • Before proceeding, check the model's discussion page to ensure no PR for base_model metadata has been created recently

4. Investigate the Base Model

   • Review the model card and repository for information about the original architecture
   • Look for mentions of base models in the model description or training details
   • If unclear, you may need to do some research on the model's origin

5. Add the Base Model Metadata (if applicable)

   • Open the model's README.md file on the Hub
   • Add or update the following in the YAML metadata section:

     base_model: HuggingFaceH4/zephyr-7b-beta
     

   • For models derived from multiple base models, use a list:

     base_model:
     - Endevor/InfinityRP-v1-7B
     - l3utterfly/mistral-7b-v0.1-layla-v4
     

   • Optionally, specify the relationship type:

     base_model_relation: quantized
     

     (Valid options: "adapter", "merge", "quantized", "fine-tune")

6. Open a Pull Request (if needed)

   • If you've added or updated metadata, submit your changes as a pull request on the model's repository
   • In the PR description, explain your reasoning for adding the base_model metadata

7. Update the Sprint CSV

   • Update the row for the model you processed:
     • If you created a PR: Set PR_CREATED to "true" and add the PR link to the PR_LINK column
     • If no PR was needed: Set PR_CREATED to "Not Required"
   • Create a pull request to update the Sprint CSV with your changes

Guidelines

• Focus on accuracy. It's better to mark "Not Required" than to add incorrect metadata.
• If you're unsure about a base model, you can open a discussion on the model's page to ask the author or community.
• For models with multiple potential base models, prioritize the most direct ancestor.
• Remember that some models may not have a clear base model or may not require this metadata. It's okay to mark these as "Not Required".

Resources

• Guide to Model Cards on Hugging Face
• Base Model Metadata Documentation
• Sprint CSV: https://huggingface.co/datasets/librarian-bots/base_model_sprint/blob/main/no_base_model_specified_has_readme_and_chat_template.csv

Let's work together to make the Hugging Face Hub an even more valuable resource for the AI community!