License, provenance and attribution for revision 2575352d61be1bf7225cf8f8b268a4678025fc58

#2
by healingsoftly - opened

Hello maintainers,

Could you clarify the provenance and redistribution terms for onnx-community/whisper-tiny.en at revision 2575352d61be1bf7225cf8f8b268a4678025fc58?

The intended use is self-hosting pinned assets behind private HTTPS access for a browser-based test, with speech inference entirely on the test device.

  1. Which upstream model revision and conversion/quantization tools or recipe produced these assets? Were the weights only converted/quantized, or otherwise modified or fine-tuned?
  2. Which license applies to the converted ONNX files, including onnx/encoder_model_quantized.onnx and onnx/decoder_model_merged_quantized.onnx, and the associated model/tokenizer/configuration files? Does it permit copying and serving them from a separate private test host?
  3. What copyright statements, attribution, LICENSE and NOTICE files must accompany those assets? The card names openai/whisper-tiny.en as the base, but the pinned conversion card does not state a license. The original Whisper repository identifies MIT, while the Hugging Face base-model card identifies Apache-2.0; please clarify the appropriate terms/notices for this converted revision and any conversion-specific material.

A link to authoritative license/provenance documentation, or an explicit repository clarification, would be appreciated. We have not treated download availability as redistribution permission.

Thank you.

ONNX Community org

Apache-2.0

Xenova changed discussion status to closed

Sign up or log in to comment