CuriousDragon/ettin-17m-sponsor-combined-android

Ettin 17M sponsor detector (Android INT8)

ONNX Runtime package used by Flow for on-device sponsor-span detection. Quantized from CuriousDragon/ettin-17m-sponsor-combined (embedding_int8_per_channel, ORT format, opset 18).

• Primary runtime file: sponsor_detector_combined.int8.ort (~28 MB)
• Also included: FP32 ONNX and INT8 ONNX for desktop/debug
• Tokenizer: same Ettin / ModernBERT tokenizer as the PyTorch checkpoints
• Confidence threshold in the Android package: 0.7

Files

File	Role
sponsor_detector_combined.int8.ort	Shipping Android model
sponsor_detector_combined.int8.onnx	INT8 ONNX before ORT packing
sponsor_detector_combined.fp32.onnx	Unquantized ONNX
tokenizer.json / tokenizer_config.json	Tokenizer
required_operators_and_types.config	Reduced ORT operator config
tokenizer_goldens.json	Tokenization goldens
sponsor_feedback_v1.schema.json	Local training-journal schema
android_manifest.json	Export manifest and hashes

ScriptSmith corpus

This model additionally used the ScriptSmith/sponsorblock-youtube-metadata-2024 transcript/metadata dataset (CC BY 4.0 compilation; its YouTube auto-captions remain under YouTube's terms). Those transcripts are rolling auto-captions that were reconstructed before alignment, so caption boundaries are approximate.

Limitations

• English transcripts only.
• Target is external paid sponsorships. Self-promotion, memberships, merchandise, and interaction reminders are treated as negatives.
• SponsorBlock labels are noisy and incomplete. Evaluations against them are weak labels, not ground truth.
• Full-video reconstruction interpolates timestamps inside caption cues.
• Window-level scores do not by themselves skip video; Flow stitches overlapping 768-token windows with a 128-token overlap.

License

These weights are released under CC BY-NC-SA 4.0.

The base encoder jhu-clsp/ettin-encoder-17m is MIT. Fine-tuning used SponsorBlock labels (CC BY-NC-SA 4.0) and the pinned Xenova/sponsorblock-768 window corpus. Non-commercial use and share-alike apply to this derivative.

This project is not affiliated with SponsorBlock.