Llama-3.2-1B-Instruct-Capybara-abliterated
joelewing/Llama-3.2-1B-Instruct-Capybara-abliterated
hereticSAFETENSORS
Published metrics
- KL divergence
- 0.030
- Refusals
- 3/100
- Original refusals
- 77/100
- Heretic version
- v1.0.1
Published on the model card
Matched snippet
| KL divergence | 0.03 | | Refusals | 3/100 |
Provenance
- Canonical model
- joelewing/Llama-3.2-1B-Instruct-Capybara
- Declared parent(s)
- joelewing/Llama-3.2-1B-Instruct-Capybara, unsloth/llama-3.2-1b-instruct-bnb-4bit
- Classification
- card states the weights were abliterated with Heretic
- Created / updated
- 2025-11-17 / 2025-11-17
- Downloads
- 141 all time · 23 last 30 days
- License
- llama3.2
Refusal rate 3%. Metrics are self-reported by the publisher; the directory does not re-run evaluations.