How this works

Loadcheck resolves a public Hugging Face repository to a commit, reads its tree and at most two small text files, and reports covered signals.

Signals

For models, it checks confirmed auto_map custom-code mappings; current Transformers requires an explicit trust_remote_code=True opt-in before such code runs. It also reports pickle-capable checkpoint paths without any .safetensors file. For datasets, it reports a root loading script only as legacy Datasets v3.4 behavior, where repository code ran only with explicit load_dataset(..., trust_remote_code=True). Template-like delimiters below README configs are neutral manual-review signals: current YAML builder parameters work without custom code, and no current standard Hugging Face loading call is documented to render them.

Limits

It does not download weights, import code, render templates, inspect arbitrary files, or certify a repository. API, parsing, size, pagination, rate-limit, or browser network failures produce an incomplete result.

Transformers supported models · Transformers Auto classes · Legacy Datasets v3.4 scripts · Dataset manual configuration · Hugging Face pickle scanning documentation