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.
├── 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.
