Get the app
← back to catalog · registered 2026-09-11 04:55

j4ys0n/MiniCPM5-2B-heretic-abliterated-ONNX

j4ys0n 2B second-order
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
M1 Direct removal →
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)
How we label → About M3 →
Refusal direction extracted via
Extraction technique

Difference-of-means

FromArditi et al. (2024) ↗
Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
What is a refusal direction? → Read the paper ↗
Downloads · 30-day
72
Likes
0
Model age
today
created 2026-09-11

Genealogy 0 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

insraq/MiniCPM5-2B-heretic-abliterated↓ 719 this lineage
j4ys0n/MiniCPM5-2B-heretic-abliterated-ONNX↓ 72

Metadata

License
apache-2.0
Languages
en zh
Tags
transformers.js onnx llama text-generation webgpu minicpm5 heretic conversational en zh base_model:insraq/MiniCPM5-2B-heretic-abliterated base_model:quantized:insraq/MiniCPM5-2B-heretic-abliterated

Related

Same base model
mradermacher/MiniCPM5-2B-heretic-abliterated-i1-GGUF 2K mradermacher/MiniCPM5-2B-heretic-abliterated-GGUF 2K bumblebuttpow/MiniCPM5-2B-heretic-abliterated-MLX-8bit 9
Total size
0 B
Files
8
Quantizations
1
Registered
2026-09-11 04:55
Last updated on HF
2026-09-11 04:55
Source huggingface.co/j4ys0n/MiniCPM5-2B-heretic-abliterated-ONNX ↗ Base model insraq/MiniCPM5-2B-heretic-abliterated →

Files by quantization

Auxiliary files 8 files 9.46 MB
tokenizer.json 9.44 MB d15448ea download
tokenizer_config.json 9.48 KB bbf0ade1 download
chat_template.jinja 8.78 KB 217bb497 download
conversion-manifest.json 1.98 KB 4f4bd56d download
.gitattributes 1.54 KB 3d92c672 download
config.json 908 B 0b78c923 download
README.md 907 B df3c4a67 download
generation_config.json 227 B 3c5ce688 download

README current version from Hugging Face

license: apache-2.0
base_model: insraq/MiniCPM5-2B-heretic-abliterated
library_name: transformers.js
pipeline_tag: text-generation
tags: [onnx, webgpu, minicpm5, heretic]
language: [en, zh]

MiniCPM5 2B Heretic — ONNX q4f16

Converted from insraq/MiniCPM5-2B-heretic-abliterated at revision 8d83eaf42e30f4e4f164b8685be0ecdccd8ef0d8. Original model: OpenBMB MiniCPM5-2B. This is the original safetensors checkpoint referenced by Abiray's GGUF model card; no GGUF files are used. Conversion changes precision, not the fine-tuning method.

Load with Transformers.js 4.2.0, device: 'webgpu', dtype: 'q4f16'. Requires shader-f16; weights are approximately 1.83 GB. Use tokenizer_encode_kwargs: { enable_thinking: false } to disable thinking.

See conversion-manifest.json for source revision, tool versions and file checksums.

README history 1 version

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-09-11Publish browser-tested MiniCPM5 Heretic q4f16 ONNX conversion41d5a26907 B
    Loading...
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.

See supported apps Browse the catalog