Heretic Directory

Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8

DavidAU/Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8

Original model on Hugging Face
30B · 3B activeQwen 33 Heretic artifacts1 measured conversions11.9K downloads
Best version98.9Pareto
ayunami2000/Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8-heretic
by ayunami2000 · Heretic v1.2.0 · AWQ, EXL, GGUF, GPTQ, SAFETENSORS
KL divergence
0.0014
Refusals
2/100
Refusal rate
2%
Downloads
11.9K
Evidence from the model card
| KL divergence | 0.0014 | | Refusals | 2/100 |

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
1ayunami200030B · 3B active0.00142/1002%98.9
BestParetoAWQEXLGGUF
11.9K

All artifacts (3)

RepositoryTypeFormatsMetricsDownloadsUpdated
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8-heretic-i1-GGUF
mradermacher
quant
GGUFIQ1_M
inheritedKL 0.001410K2026-03-05
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8-heretic-GGUF
mradermacher
quant
GGUFIQ4_XS
inheritedKL 0.00141.8K2026-03-05
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-HI8-heretic
ayunami2000
heretic
SAFETENSORS
published (weak match)KL 0.00141032026-03-04