Heretic Directory

medgemma-27b-text-it

google/medgemma-27b-text-it

Original model on Hugging Face
27BGemma 26 Heretic artifacts1 measured conversions3.1K downloads
Best version84.1Pareto
pszemraj/medgemma-27b-text-heretic_med
by pszemraj · Heretic v1.0.1 · GGUF, MLX, SAFETENSORS
KL divergence
0.050
Refusals
27/100
Refusal rate
27%
Downloads
3.1K
Evidence from the model card
| KL divergence | 0.05 | | Refusals | 27/100 |

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
1pszemraj27B0.05027/10027%84.1
BestParetoGGUFMLXSAFETENSORS
3.1K

All artifacts (6)

RepositoryTypeFormatsMetricsDownloadsUpdated
medgemma-27b-text-heretic_med-mlx-6Bit
Wwayu
quant
MLXSAFETENSORS
inheritedKL 0.0509132025-12-15
medgemma-27b-text-heretic_med-Q4_K_M-GGUF
boroll2347
quant
GGUFQ4_K_M
inheritedKL 0.0504912026-05-17
medgemma-27b-text-heretic_med
pszemraj
heretic
SAFETENSORS
published (weak match)KL 0.0504722025-12-29
medgemma-27b-text-heretic_med-Q8_0-GGUF
boroll2347
quant
GGUFQ8_0
inheritedKL 0.0504472026-05-17
medgemma-27b-text-heretic_med-Q5_K_M-GGUF
SzilviaB
quant
GGUFQ5_K_M
inheritedKL 0.0504322026-07-02
medgemma-27b-text-heretic_med-Q2_K-GGUF
boroll2347
quant
GGUFQ2_K
inheritedKL 0.0503442026-05-17