← back to catalog · registered 2026-09-11 12:55

dgvcxz/MiniCPM5-2B-heretic-abliterated-int4-gq-ov

Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
Downloads · 30-day
0
Likes
0
Model age
today
created 2026-09-11

Metadata

License
apache-2.0
Tags
openvino llama license:apache-2.0 region:us

Related

Total size
1.46 GB
Files
14
Quantizations
1
Registered
2026-09-11 12:55
Last updated on HF
2026-09-11 12:17

Files by quantization

Auxiliary files 14 files 1.47 GB
openvino_model.bin 1.45 GB 8386cced download
openvino_tokenizer.bin 4.65 MB c1563549 download
openvino_detokenizer.bin 1.83 MB 013bc707 download
tokenizer.json 9.44 MB d15448ea download
openvino_model.xml 3.30 MB 8071a936 download
openvino_tokenizer.xml 33.8 KB 7ae710e2 download
openvino_detokenizer.xml 17.3 KB c6842831 download
chat_template.jinja 8.85 KB be166eec download
.gitattributes 1.48 KB a6344aac download
config.json 748 B 54b173a1 download
tokenizer_config.json 540 B 5fdb6787 download
openvino_config.json 471 B 5b12f95e download
README.md 231 B 49234335 download
generation_config.json 213 B 0ec6d84f download

README current version from Hugging Face


license: apache-2.0

optimum-cli export openvino -m insraq/MiniCPM5-2B-heretic-abliterated --task text-generation-with-past --weight-format int4 --sym --ratio 1.0 --group-size 128 MiniCPM5-2B-heretic-abliterated-int4-gq-ov

Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in app" button that hands off directly to a local runtime of your choice - Infrahuman, LM Studio, or Ollama. No API keys, no subscription, no prompt leakage.