hf_readmes

HuggingFace model card sources

One folder per HuggingFace model repo under christopherthompson81. Each <repo-name>/README.md here is the source of truth for that HF repo's model card; the file pushed to HF should match the file in this directory.

Repo map

HF repo	Source folder	License	Status
sortformer_parakeet_onnx	sortformer_parakeet_onnx/	mixed (CC-BY-4.0 / NVIDIA OML / MIT / Apache-2.0)	needs upload
nemotron3_diarization_onnx	nemotron3_diarization_onnx/	OpenMDW-1.1	live
diarizen_onnx	diarizen_onnx/	CC-BY-NC-4.0	needs upload (current repo has no README)
cohere-transcribe-03-2026-onnx	cohere-transcribe-03-2026-onnx/	Apache-2.0	needs upload
vibevoice-asr-onnx	vibevoice-asr-onnx/	MIT	needs upload
voxlingua107-lid-onnx	voxlingua107-lid-onnx/	Apache-2.0	needs upload
indicconformer-600m-onnx	indicconformer-600m-onnx/	MIT	needs upload
granite-speech-4-1-2b-onnx	granite-speech-4-1-2b-onnx/	Apache-2.0	needs upload
granite-speech-4-1-2b-onnx-bf16	granite-speech-4-1-2b-onnx-bf16/	Apache-2.0	needs upload
omnivoice-ipa-onnx	omnivoice-ipa-onnx/	per file: CC-BY-NC-4.0 (transformer, diff) / Boson Higgs Audio 2 Community (codec) / Apache-2.0 (tokenizer) / per-clip CC-BY, CC0, CC-BY-SA, PD (voice library)	live
kokoro-82m-onnx	kokoro-82m-onnx/	Apache-2.0	live — see scripts/kokoro_export/README.md
chatterbox-tts-onnx	chatterbox-tts-onnx/	MIT	live — see scripts/chatterbox_export/README.md

What each card includes

• YAML frontmatter — license, license_link / license_name for other, library_name: onnxruntime, pipeline_tag, tags, base_model, language
• A header block with the upstream link and a one-line "for use with Vernacula" statement
• Contents table — every file in the bundle, what it does, which upstream component it derives from
• Export provenance — link to the scripts/<name>_export/ folder on GitHub and a short paragraph on what the export does
• License section — explicit, with per-component breakdown for mixed-license bundles
• Using these files — minimal huggingface_hub + onnxruntime snippet for the non-Vernacula path
• Limitations — pointer to upstream model card with a note on what's specific to the ONNX repackaging
• Citation — BibTeX where I have it, link to upstream otherwise
• Acknowledgments — credits original authors, names the repackager
• See also — Vernacula repo, conversion script, upstream model card, the rest of the namespace

Uploading

The canonical path is scripts/upload_to_hf.py, which auto-resolves the README from this directory by repo basename:

# Sync the README only (no model artifacts)
python scripts/upload_to_hf.py \
    --model-dir /tmp/empty \
    --repo-id christopherthompson81/<repo-name> \
    --files /dev/null \
    --sync-readme

# Or full bundle + manifest + README
python scripts/make_manifest.py --model-dir ~/models/<bundle> --all
python scripts/upload_to_hf.py \
    --model-dir ~/models/<bundle> \
    --repo-id christopherthompson81/<repo-name> \
    --sync-readme --create-repo

For a quick README-only sync, the underlying huggingface-cli call also works:

huggingface-cli upload christopherthompson81/<repo-name> \
    scripts/hf_readmes/<repo-name>/README.md README.md \
    --commit-message "sync model card from vernacula repo"

When updating a card, edit the file here, commit, and re-upload — never hand-edit the README on the Hub directly, since changes there will be silently overwritten on the next sync.