DominicTWHV/Cockatoo-Moderation-V1
Viewer β’ Updated β’ 276k β’ 140 β’ 2
This is a preliminary test model and is NOT meant for production use.
| Base model | microsoft/deberta-v3-small |
| Dataset | Cockatoo Moderation V1 (276k annotated subset) |
| Epochs | 3 |
| Task | Text classification (moderation) |
| Language | English |
| License | CC-BY-4.0 |
Starting from microsoft/deberta-v3-small and training for 3 epochs, this model evaluates the usability of the 276k annotated subset of the Cockatoo Moderation V1 dataset.
The
harassment_targetlabel in the model's output is an artifact of the input training data and should be ignored.
The exact training data and splits are available in the training-data/ folder.
See training_params.json.
Cockatoo Data Studio (annotation)
β
Cockatoo ML Orchestrator
β
Cockatoo ML Worker (training agent)
Training completed in ~1 hour on 2Γ NVIDIA P100 GPUs (over PCIe).
Base model
microsoft/deberta-v3-small