DarrenJiaImbue/ai-detection-demo-gemma-logits
ai-detection-demo — Gemma 4 E4B logits Pre-computed last-token vocabulary logits from int4-quantized google/gemma-4-E4B-it, applied over DarrenJiaImbue/ai-detection-demo-dataset. These logits are the training input to DarrenJiaImbue/ai-detection-demo-gemma_4_e4b, the linear-probe classification head. Structure ├── train/ │ ├── logits.npy (14000, 262144) fp16 │ ├── labels.npy (14000,) int64 — cosine-distance bucket 0..3 │ └── meta.json ├── val/ │… See the full description on the dataset page: https://huggingface.co/datasets/DarrenJiaImbue/ai-detection-demo-gemma-logits.
Card: drop iOS/Bouncer/LiteRT references; fix bucket boundaries
Card: drop undo_softcap example, clarify logits are used as-is
Rename google_reviews -> yelp_reviews in test/sample.parquet
Rename google_reviews -> yelp_reviews in val/sample.parquet
Initial: Mac Metal Gemma 4 E4B logits over ai-detection-demo-dataset (train/val/test)
initial commit