asaverren/chat-format-writer

Chat Format Writer

Peek at a Hub model's real tokenizer/template files, hash the live chat template, and queue a verified append to `asaverren/chat-format-index`.

This Space is Static

Free new accounts cannot create Gradio or Docker Spaces (402). This page is plain HTML: it fetches public Hub tokenizer/template files in the browser and opens a dataset discussion. It does not run Python, download weights, or write the dataset itself.

Local Gradio / CLI (real verifier)

Clone this Space and run the same quality bar that writes the index:

python -m venv .venv && . .venv/bin/activate
pip install -r requirements.txt
python ingest.py org/name          # dry-run
python ingest.py org/name --write  # append a passing row (needs HF_TOKEN)
python app.py                      # local Gradio UI

No token lives in these files. HF_TOKEN is read from the environment at runtime. trust_remote_code=False. Weights are never downloaded.

Queue a write from the Static Space

1. 1.Preview a model_id here (client-side, public Hub only).
2. 2.Request add opens a new discussion on the dataset.
3. 3.Title it exactly add: org/name (the page tries to prefill this).
4. 4.Someone with write access runs ingest.py — tokenizer/template only. Duplicates and invented templates are refused.

Quality bar

• —Real jinja/template on Hub files, or we refuse. No invented templates.
• —Tokenizer/template files only. Never weights.
• —trust_remote_code=False.
• —Duplicate model_id already in the index is a refuse, not an overwrite.