Heretic Directory

LFM2.5-8B-A1B

LiquidAI/LFM2.5-8B-A1B

Original model on Hugging Face
8B · 1B activeLFM10 Heretic artifacts3 measured conversions31.2K downloads
Best version97.7Pareto
zaakirio/LFM2.5-8B-A1B-Uncensored
by zaakirio · GGUF, SAFETENSORS
KL divergence
0.048
Refusals
0/100
Refusal rate
0%
Downloads
8.5K
Evidence from the model card
| Refusals (/100 harmful prompts) | 0 | | KL divergence (harmless prompts) | 0.0481 |

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
1zaakirio8B · 1B active0.0480/1000%97.7
BestParetoGGUFSAFETENSORS
8.5K
2coder31018B · 1B active0.02412/10012%92.8
ParetoGGUFMLXSAFETENSORS
20.5K
3saidutta698B · 1B active0.16528/10028%78.9
GGUFSAFETENSORS
2.2K

All artifacts (10)

RepositoryTypeFormatsMetricsDownloadsUpdated
LFM2.5-8B-A1B-heretic-i1-GGUF
mradermacher
quant
GGUFIQ1_M
inheritedKL 0.0249.1K2026-06-07
LFM2.5-8B-A1B-Uncensored-GGUF
zaakirio
quant
GGUFBF16
published (weak match)KL 0.0487.3K2026-06-04
LFM2.5-8B-A1B-heretic-GGUF
mradermacher
quant
GGUFIQ4_XS
inheritedKL 0.0244.9K2026-06-07
LFM2.5-8B-A1B-heretic-imatrix-GGUF
FadedRedStar
quant
GGUFIQ4_NL
published (weak match)KL 0.0243.3K2026-07-10
LFM2.5-8B-A1B-heretic
saidutta69
heretic
GGUFSAFETENSORSF16
published (weak match)KL 0.1652.2K2026-10-05
LFM2.5-8B-A1B-heretic-GGUF
FadedRedStar
quant
GGUFQ4_K_M
published (weak match)KL 0.0241.9K2026-07-10
LFM2.5-8B-A1B-Uncensored-GGUF
lopezjhonf
quant
GGUFQ6_K
published (weak match)KL 0.0481K2026-07-01
LFM2.5-8B-A1B-heretic-16e-GGUF
texassred
quant
GGUFSAFETENSORS
inheritedKL 0.0249772026-08-17
LFM2.5-8B-A1B-heretic
coder3101
heretic
SAFETENSORS
published (weak match)KL 0.0242652026-06-06
LFM2.5-8B-A1B-Uncensored
zaakirio
heretic
SAFETENSORS
published (weak match)KL 0.0482062026-06-04