Heretic Directory

SmolLM3-3B

HuggingFaceTB/SmolLM3-3B

Original model on Hugging Face
3BSmolLM13 Heretic artifacts3 measured conversions6.7K downloads
Best version70.0Pareto
MassivDash/SmolLM3-3B-heretic
by MassivDash · Heretic v1.1.0 · GGUF, SAFETENSORS
KL divergence
0.0001
Refusals
60/100
Refusal rate
60%
Downloads
4.8K
Evidence from the model card
| KL divergence | 0.0001 | | Refusals | 60/100 |

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
1MassivDash3B0.000160/10060%70.0
BestParetoGGUFSAFETENSORS
4.8K
2xtools-at3B0.0494/1004%95.7
unverifiedParetoGGUF
604
3rainmana3B065/10065%67.5
unverifiedimplausibleParetoSAFETENSORS
187

All artifacts (13)

RepositoryTypeFormatsMetricsDownloadsUpdated
SmolLM3-3B-heretic-i1-GGUF
mradermacher
quant
GGUFIQ1_M
inheritedKL 0.00012.6K2026-01-09
SmolLM3-3B-heretic-GGUF
mradermacher
quant
GGUFIQ4_XS
inheritedKL 0.00011.7K2026-01-09
SmolLM3-3B-Instruct-heretic-GGUF
xtools-at
quant
GGUFQ2_K_L
published (weak match)KL 0.0496042026-09-19
SmolMed3-3B-GGUF
xtools-at
finetune
GGUFQ2_K_L
none4292026-10-12
SmolLM3-3B-heretic
MassivDash
heretic
SAFETENSORS
published (weak match)KL 0.00013052026-01-08
SmolLM3-3B-Rebel
XuehangCang
quant
GGUFSAFETENSORSQ4_K_M
noneKL 02982026-06-29
SmolLM3-3B-Medical-Reasoning-GGUF
xtools-at
finetune
GGUFQ4_K_M
none1922026-10-14
SmolLM3-3B-heretic
rainmana
heretic
SAFETENSORS
published (weak match)KL 01872026-01-19
SmolPhish3-3B-GGUF
xtools-at
finetune
GGUFQ4_K_M
none1702026-09-19
SmolLM3-3B-heretic-Q8_0-GGUF
MassivDash
quant
GGUFQ8_0
inheritedKL 0.00011662026-01-08
SmolLM3-3B-Medical-Reasoning
xtools-at
finetune
LORASAFETENSORS
none02026-10-14
SmolMed3-3B
xtools-at
finetune
LORASAFETENSORS
none02026-10-12
SmolPhish3-3B
xtools-at
finetune
LORASAFETENSORS
none02026-09-19

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