| 1 | Owl Qwen — single-neuron deltas |
| 2 | |
| 3 | Training data: Sauers/Qwen3-8B-tawny-owl-128-neurons-data |
| 4 | Revision: f6264579102b8ebeb3d6fa418c184b686a96af8c |
| 5 | https://huggingface.co/datasets/Sauers/Qwen3-8B-tawny-owl-128-neurons-data |
| 6 | Dataset license: CC BY-NC 4.0 |
| 7 | https://creativecommons.org/licenses/by-nc/4.0/ |
| 8 | |
| 9 | Base checkpoint: daavidhauser/Swift-1.5-Qwen3.8-27B-W4A16-HyperQwen-INT4 |
| 10 | Revision: 64515d381e36084182e4e2abfa20050b75f29889 |
| 11 | |
| 12 | owl.safetensors contains only learned gate/up/down deltas for one MLP neuron. |
| 13 | It contains no base-model weights or personal training data. The checkpoint, |
| 14 | layer, neuron and exact base-manifest fingerprint are recorded in its metadata. |
| 15 | The experiment and validation record is in tools/owl-neuron/README.md. |