Heretic Directory

fable-qwen2.5-3b-agentic-merged

devxyasir/fable-qwen2.5-3b-agentic-merged

Original model on Hugging Face
3BQwen 2.53 Heretic artifacts1 measured conversions4.6K downloads
Best version96.1Pareto
richardyoung/fable-qwen2.5-3b-abliterated-mlx
by richardyoung · MLX, SAFETENSORS
KL divergence
0.050
Refusals
3/100
Refusal rate
3%
Downloads
676
Evidence from the model card
| Refusals after | 3 / 100 | | KL divergence | 0.050 |

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
1richardyoung3B0.0503/1003%96.1
BestParetoMLXSAFETENSORS
676

All artifacts (3)

RepositoryTypeFormatsMetricsDownloadsUpdated
fable-qwen2.5-3b-agentic-merged-heretic-i1-GGUF
mradermacher
quant
GGUFIQ1_M
none2.9K2026-06-27
fable-qwen2.5-3b-agentic-merged-heretic-GGUF
mradermacher
quant
GGUFIQ4_XS
none9982026-06-27
fable-qwen2.5-3b-abliterated-mlx
richardyoung
quant
MLXSAFETENSORS
published (weak match)KL 0.0506762026-06-26

2 artifacts inherit metrics from their parent conversion or publish none.