[RFC] HuggingFace Community Model Guidelines

[byjlw]

These are the proposed guidelines for models uploaded to Hugging Face Community

One thing that's missing that we should nail down is a standard naming for quant schemes. Right now things aren't standardized enough for people to infer things. GGUF does a good job at this.

Guidelines

• Each high level item will target a particular model and target device
• Model card will include
  • Boiler Plate Details on ET
  • Link to the original model
  • License or link to license
  • Recipe for converting the original model to all the quantized versions
  • Code snip-it on how to use the model
  • Prompt format (if LLM)
  • Table for all models within this repo with size and expected results along as a recommendation on which one to use
    • FileName
    • QuantType
    • Size
    • Description
  • Besides .pte file, Corresponding .pt2 file and its recipe (Future)
• Files will include
  • .gitattributes
  • .pte files for all the quantized versions of the model
  • config.json
  • export.ipynb
  • README.md
  • params.json
  • tokenizer.model

Repo Naming

Names should include model, number of parameters and target device
{Model}-{Params}-{target-device}
Example: executorch-community/Llama-3.2-1B-Instruct-ARM-CPU

Model File Naming

{Model}-{Params}-{quantscheme}-{bitsize}.pte
Example: llama-3.2-1B-SQ-4-8.pte

Quant Lookup

General format
{Scheme as Abbreviation}-{weight bit size}-{activation bit size}

[mergennachin]

Also, attach appropriate licenses as well as possibly a legal disclaimer regarding quantization - something similar to https://github.com/pytorch/executorch/tree/main/examples#disclaimer

cc @orionr, @supriyar, @gchanan

[byjlw]

updated the guide