Heretic Directory

Qwen2.5-32B-Instruct

Qwen/Qwen2.5-32B-Instruct

Original model on Hugging Face
32BQwen 2.53 Heretic artifacts1 measured conversions246 downloads
Best version93.7Pareto
dec0dedd/Qwen2.5-32B-Instruct-heretic
by dec0dedd · Heretic v1.2.0 · GGUF, SAFETENSORS
KL divergence
0.081
Refusals
5/100
Refusal rate
5%
Downloads
246
Evidence from the model card
| KL divergence | 0.0812 | | Refusals | 5/100 |

Metrics published by the author. Why this ranks first.

All measured versions

Higher score is better. One row per ablation attempt; artifacts of the same attempt are grouped under it.

#AuthorParamsKLRefusalsScoreEvidenceDownloads
1dec0dedd32B0.0815/1005%93.7
BestParetoGGUFSAFETENSORS
246

All artifacts (3)

RepositoryTypeFormatsMetricsDownloadsUpdated
Qwen2.5-32B-Instruct-heretic-Q8_0-GGUF
QrCode99
quant
GGUFQ8_0
inheritedKL 0.0811572026-03-21
Qwen2.5-32B-Instruct-heretic
dec0dedd
heretic
SAFETENSORS
published (weak match)KL 0.081892026-03-19
Qwen2.5-32B-Instruct-abliterated
varun-balani
heretic
GGUFSAFETENSORSQ4_K_M
none02026-10-17

1 artifact inherit metrics from their parent conversion or publish none.