Heretic Directory

Qwen2.5-14B-Instruct-1M

Qwen/Qwen2.5-14B-Instruct-1M

Original model on Hugging Face
14BQwen 2.51 Heretic artifacts1 measured conversions624 downloads
Best version87.6Pareto
richardyoung/Qwen2.5-14B-Instruct-1M-heretic
by richardyoung · Heretic v1.0.1 · SAFETENSORS
KL divergence
0.280
Refusals
3/100
Refusal rate
3%
Downloads
624
Evidence from the model card
| KL divergence | 0.28 | | Refusals | 3/100 |

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
1richardyoung14B0.2803/1003%87.6
BestParetoSAFETENSORS
624

All artifacts (1)

RepositoryTypeFormatsMetricsDownloadsUpdated
Qwen2.5-14B-Instruct-1M-heretic
richardyoung
heretic
SAFETENSORS
published (weak match)KL 0.2806242026-09-26