Heretic Directory

Qwen3.8-27B-Ultra-Uncensored-Heretic-Native-MTP-Preserved

llmfan46/Qwen3.8-27B-Ultra-Uncensored-Heretic-Native-MTP-Preserved

Original model on Hugging Face
27BQwen 3.89 Heretic artifacts1 measured conversions124.1K downloads
Best version97.3Pareto
llmfan46/Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved-GGUF
by llmfan46 · Heretic v2.0.0 · GGUF, GPTQ, NVFP4, SAFETENSORS
KL divergence
0.024
Refusals
3/100
Refusal rate
3%
Downloads
115.7K
Evidence from the model card
| KL divergence | 0.0244 | | Refusals | 3/100 |

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
1llmfan4627B0.0243/1003%97.3
BestParetoGGUFGPTQNVFP4
115.7K

All artifacts (9)

RepositoryTypeFormatsMetricsDownloadsUpdated
Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved-GGUF
llmfan46
quant
GGUFBF16
published (weak match)KL 0.02476.3K2026-08-31
Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved-NVFP4-GGUF
llmfan46
quant
GGUFNVFP4NVFP4
published (weak match)KL 0.02424.9K2026-08-31
Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved-GPTQ-Int4
llmfan46
quant
GPTQSAFETENSORSINT4
published (weak match)KL 0.0246.8K2026-08-31
Qwen3.8-27B-Uncensored-Heretic-Native-MTP-Preserved-NVFP4
llmfan46
quant
NVFP4SAFETENSORSNVFP4
published (weak match)KL 0.0246.5K2026-08-31
Qwen3.8-27B-Heretic-Dynamic-GGUF
ajgazin
quant
GGUFQ4_K_XL
none6K2026-09-01
Qwen3.8-27B-Heretic-NVFP4-NInfer
koloved
finetune
NVFP4NVFP4
none2.4K2026-09-07
Qwen3.8-27B-Ultra-Uncensored-Heretic-Native-MTP-Preserved-GGUF
Lalo1999
quant
GGUFBF16
published (weak match)KL 0.0241.2K2026-09-01
Qwen3.8-27B-Ultra-Uncensored-Heretic-Native-MTP-Preserved-GGUF-llamafile
surfrpt
quant
GGUF
none932026-09-11
Qwen3.8-27B-Heretic-MTP-Q27
mmlkk
finetune
OTHER
none02026-09-21

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