Heretic Directory

SmolLM3-3B

unsloth/SmolLM3-3B

Original model on Hugging Face
3BSmolLM1 Heretic artifacts1 measured conversions93 downloads
Best version85.5Pareto
DrRiceIO7/SmolLM3-3B-heretic
by DrRiceIO7 · Heretic v1.1.0 · GGUF, SAFETENSORS
KL divergence
0.0002
Refusals
29/100
Refusal rate
29%
Downloads
93
Evidence from the model card
| KL divergence | 0.0002 | | Refusals | 29/100 |

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
1DrRiceIO73B0.000229/10029%85.5
BestParetoGGUFSAFETENSORS
93

All artifacts (1)

RepositoryTypeFormatsMetricsDownloadsUpdated
SmolLM3-3B-heretic
DrRiceIO7
heretic
SAFETENSORS
published (weak match)KL 0.0002932026-01-21