Heretic Directory

Llama-3.3-70B-Instruct

meta-llama/Llama-3.3-70B-Instruct

Original model on Hugging Face
70BLlama 38 Heretic artifacts3 measured conversions125.2K downloads
Best version66.5Pareto
TeeZee/Llama-3.3-70B-Instruct-heretic
by TeeZee · Heretic v1.0.1 · SAFETENSORS
KL divergence
1.63
Refusals
5/100
Refusal rate
5%
Downloads
349
Evidence from the model card
| KL divergence | 1.63 | | Refusals | 5/100 |

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
1TeeZee70B1.635/1005%66.5
BestParetoSAFETENSORS
349
2kldzj70B1.856/1006%64.5
AWQGGUFSAFETENSORS
107K
3TeeZee70B2.413/1003%63.2
ParetoGGUFSAFETENSORS
17.9K

All artifacts (8)

RepositoryTypeFormatsMetricsDownloadsUpdated
Llama-3.3-70B-Instruct-heretic-i1-GGUF
mradermacher
quant
GGUFIQ1_M
inheritedKL 1.8595.3K2026-05-17
Llama-3.3-70B-Instruct-heretic-v2-i1-GGUF
mradermacher
quant
GGUFIQ1_M
inheritedKL 2.4113.9K2025-12-07
Llama-3.3-70B-Instruct-heretic-GGUF
mradermacher
quant
GGUFIQ4_XS
inheritedKL 1.859.5K2026-10-13
Llama-3.3-70B-Instruct-heretic-v2-GGUF
mradermacher
quant
GGUFIQ4_XS
inheritedKL 2.413.5K2025-12-06
Llama-3.3-70B-Instruct-heretic-awq
kldzj
quant
AWQSAFETENSORS
published (weak match)KL 1.851.6K2025-12-01
Llama-3.3-70B-Instruct-heretic
kldzj
heretic
SAFETENSORS
published (weak match)KL 1.855552025-11-23
Llama-3.3-70B-Instruct-heretic-v2
TeeZee
heretic
SAFETENSORS
published (weak match)KL 2.414772025-12-05
Llama-3.3-70B-Instruct-heretic
TeeZee
heretic
SAFETENSORS
published (weak match)KL 1.633492025-12-05