Text Classification
Transformers
Safetensors
English
modernbert
refusal-classification
refusal-detection
ai-safety
ai-alignment
heretic
text-embeddings-inference
Instructions to use saidutta69/vantablack with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use saidutta69/vantablack with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="saidutta69/vantablack")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("saidutta69/vantablack") model = AutoModelForSequenceClassification.from_pretrained("saidutta69/vantablack", device_map="auto") - Notebooks
- Google Colab
- Kaggle
Commit History
4 days ago
-
Card: link the reporting discussion
Verified signatureMislabelled examples and integration bugs go to a thread with a template, so reports arrive reproducible instead of freeform.c0aae29 -
Card: state the 512-token ceiling and where it does and does not matter
Verified signatureHeretic scores a prompt plus about 100 generated tokens, so 512 is enough there. For general use the response tail is invisible to the model, which the card now says instead of leaving it to the Usage note.19625d9 -
Card: correct the Brier comparison, name the four residual errors, pin the Heretic sha
Verified signatureBrier 0.0173 is 40% below v2's 0.0287, not 'under half'. The four test errors are now described as they are. The Heretic snippet carries the real commit instead of a placeholder.980b949
5 days ago
-
Model card: comparison as charts, no dataset version label
Verified signatureReplaces the results tables with charts on a consistent basis: baselines at their calibrated behaviour, VantaBlack at its baked-in calibration, so all three are on the same footing. Adds a per-class comparison and a training curve with the selected epoch marked. Drops the two auto-generated figures that were unreadable at this value range. Removes the dataset version from the card; the version lives on the dataset.d3caa9c -
VantaBlack on dataset v5: calibrated ModernBERT-large, macro F1 0.9911
Verified signatureWeights are the s17 checkpoint trained on refusal-classification-dataset v5, with validation-fitted calibration (T=0.75, compliant bias 1.0) folded into the classifier head. Plain argmax reproduces the reported predictions exactly: 445 of 445 test rows identical, 4 errors. Replaces the 0.7924 checkpoint, and removes figures and provenance files from that older model that no longer describe these weights.dcf6697
15 days ago
-
Calibration is now baked into the published weights (see Calibration section)
Verified signaturef45165c -
Use the stat strip for the headline numbers
Verified signature5f5c441 -
Clarify teacher labels
Verified signaturecc9e129 -
Rebuild model card in branded layout with dark figures
Verified signature9401834 -
Model card: add training curve
Verified signaturec175c38 -
Model card: add headline, seed-spread, and per-class figures
Verified signature171c1c8 -
Add model card
Verified signaturefb6d2d9