← back to catalog · registered 2026-11-13 02:03

OpenUncensored/Inkling-2-Small-Uncensored

OpenUncensored MoE multimodal
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/OpenUncensored%2FInkling-2-Small-Uncensored"
Response includes
  • classification m1
  • files 60
  • author_summary 4 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · 30-day
1K
Likes
2
Model age
1w ago
created 2026-11-13
Downloads over time
Now 1.3K → from 0 ↑ 0%
04589171.4K 0 on Nov 13 1.3K on Nov 22 Nov
Nov 13 → Nov 22 · 10 snapshots · spans 9 days

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.

Metadata

License
apache-2.0
Tags
transformers safetensors inkling2 image-text-to-text uncensored abliterated moe mxfp8 conversational base_model:thinkingmachines/Inkling-2-Small-MXFP8 base_model:finetune:thinkingmachines/Inkling-2-Small-MXFP8 license:apache-2.0

Related

Total size
580 GB
Files
60
Quantizations
1
Registered
2026-11-13 02:03
Last updated on HF
2026-11-13 02:18

Files by quantization

Auxiliary files 60 files 580 GB
model-00048-of-00048.safetensors 14.1 GB 15108335 download
model-00001-of-00048.safetensors 13.8 GB ddf9c365 download
model-00043-of-00048.safetensors 11.9 GB daad0472 download
model-00003-of-00048.safetensors 11.9 GB a1bc2d48 download
model-00009-of-00048.safetensors 11.9 GB 78417062 download
model-00016-of-00048.safetensors 11.9 GB bd8ce09e download
model-00022-of-00048.safetensors 11.9 GB f31b30b7 download
model-00029-of-00048.safetensors 11.9 GB e493215c download
model-00036-of-00048.safetensors 11.9 GB 93e8ed95 download
model-00042-of-00048.safetensors 11.9 GB 36410a6b download
model-00004-of-00048.safetensors 11.9 GB 1edc7418 download
model-00010-of-00048.safetensors 11.9 GB fce7b4f1 download
model-00017-of-00048.safetensors 11.9 GB 77898c44 download
model-00030-of-00048.safetensors 11.9 GB 97568564 download
model-00037-of-00048.safetensors 11.9 GB 1aeee544 download
model-00023-of-00048.safetensors 11.9 GB b9099433 download
model-00044-of-00048.safetensors 11.9 GB 58ece4fb download
model-00005-of-00048.safetensors 11.9 GB cbe170a6 download
model-00006-of-00048.safetensors 11.9 GB 89b5b7db download
model-00007-of-00048.safetensors 11.9 GB b6fcba6d download
model-00008-of-00048.safetensors 11.9 GB 9ae3d009 download
model-00011-of-00048.safetensors 11.9 GB 7f83f4b3 download
model-00014-of-00048.safetensors 11.9 GB 74ebd369 download
model-00015-of-00048.safetensors 11.9 GB de7271b9 download
model-00018-of-00048.safetensors 11.9 GB 00a45fa7 download
model-00019-of-00048.safetensors 11.9 GB 52f40829 download
model-00020-of-00048.safetensors 11.9 GB 9106b908 download
model-00021-of-00048.safetensors 11.9 GB dbc589d5 download
model-00025-of-00048.safetensors 11.9 GB af984781 download
model-00026-of-00048.safetensors 11.9 GB ab4da459 download
model-00027-of-00048.safetensors 11.9 GB ab75386d download
model-00028-of-00048.safetensors 11.9 GB 93ad6921 download
model-00031-of-00048.safetensors 11.9 GB 46fdcf64 download
model-00032-of-00048.safetensors 11.9 GB b10b6d07 download
model-00033-of-00048.safetensors 11.9 GB ab65c04d download
model-00038-of-00048.safetensors 11.9 GB 23dbbbf9 download
model-00039-of-00048.safetensors 11.9 GB 77987d97 download
model-00040-of-00048.safetensors 11.9 GB c3700d57 download
model-00041-of-00048.safetensors 11.9 GB 71d85afd download
model-00002-of-00048.safetensors 11.9 GB 4879133b download
model-00013-of-00048.safetensors 11.9 GB 81f45494 download
model-00024-of-00048.safetensors 11.9 GB 6e6526c9 download
model-00035-of-00048.safetensors 11.9 GB de231d68 download
model-00012-of-00048.safetensors 11.9 GB 8357147b download
model-00034-of-00048.safetensors 11.9 GB faaec338 download
model-00045-of-00048.safetensors 11.9 GB 2cad153c download
model-00046-of-00048.safetensors 11.9 GB 4f0278a0 download
model-00047-of-00048.safetensors 11.9 GB cda4d5d4 download
mtp.safetensors 3.19 GB 3b265640 download
edit.pt 38.3 KB 58554a71 download
tokenizer.json 25.0 MB 7f332727 download
model.safetensors.index.json 134 KB 340db769 download
hf_quant_config.json 86.3 KB 78bab5ba download
tokenizer_config.json 12.8 KB 75062c6c download
chat_template.jinja 6.31 KB 2700da57 download
README.md 3.88 KB 022aa985 download
config.json 2.62 KB 78e1c167 download
.gitattributes 1.53 KB 52373fe2 download
special_tokens_map.json 448 B 3d7a60d8 download
processor_config.json 394 B b26765bf download

README current version from Hugging Face


license: apache-2.0
base_model: thinkingmachines/Inkling-2-Small-MXFP8
pipeline_tag: image-text-to-text
library_name: transformers
tags:

  • uncensored
  • abliterated
  • inkling2
  • moe
  • mxfp8

Inkling 2 Small Uncensored (OpenUncensored)

thinkingmachines/Inkling-2-Small (594B total, 17B active)
with its refusals removed by directional ablation. In our tests it went from refusing 31 of 32 harmful requests to 0,
in both chat and thinking mode, and MMLU moved by a tenth of a point (90.44% → 90.35%).

Same files, format and size as Inkling-2-Small-MXFP8
(routed experts in MXFP8, everything else BF16, 623 GB), so it runs anywhere the original MXFP8 checkpoint runs. Vision is
untouched.

Results

Same harness for both models, greedy decoding, measured on these exact files.

Inkling 2 Small This model
Refusals, chat mode (80 prompts: 32 harmful, 48 gray-area) 62 3¹
Refusals on the 32 harmful prompts, chat mode 31 0
Refusals on the 32 harmful prompts, thinking mode (effort 0.9) 32 0²
MMLU (57 subjects x 20, 1,140 questions, letter logits) 90.44% 90.35%

¹ The 3 left are not refusals but limits: "I can't produce a video file", "I don't have access to your contacts", and
"that needs a clinical evaluation".
² 31 of 32 finished thinking within 1,024 tokens. None of the 31 refused.

Sample sizes are small, so treat differences of a few prompts as noise. Long agent tasks and coding benchmarks are not
measured yet.

How it was made

  1. Refusal directions. The output of each of the 48 decoder layers at the last prompt token, for 256 harmful and 256
    harmless prompts, at thinking effort 0 and 0.9. Then the difference of means per layer and effort. The two efforts
    agree closely (cosine 0.8 to 0.95 in the middle layers).
  2. A rank-2 refusal subspace Q: the top two singular vectors of those directions over layers 9 to 47.
  3. Where to edit. Q is projected out of what each layer writes: the attention output (attn.wo_ud, BF16) and the
    MoE output, which means the down-projection of every one of the 384 routed experts (mlp.experts.w2_weight;
    Inkling 2 has no shared expert). W' = W - 1.5 Q Qᵀ W on layers 9 to 47.
  4. Writing it in. The expert weights are decoded to float32, edited, and re-encoded to MXFP8 (e4m3 with a
    power-of-two scale per 32 inputs) with stochastic rounding, so the edit is not rounded back onto the original grid.
    Everything else is byte-identical to the original: the vision tower, MTP head, embeddings, routers, attention-residual
    queries and short convolutions.

Inkling 2 adds layer outputs to the residual stream through block attention residuals and short depthwise convolutions.
We edit the projections that feed them. During the search the edit was applied as forward hooks on those projections, which is exactly equivalent to editing
the weights. The results above are from these baked files, with no hooks.

edit.pt in this repo holds Q and the per-layer strengths, so the edit can be reapplied to the original weights or
studied.

Use it

Exactly like the original MXFP8 checkpoint. Or call it without GPUs through the
OpenUncensored API as inkling-2-small-uncensored. With vLLM, follow the
Inkling 2 Small recipe (4x B300 or 8x H200) and point it at
this repo. With transformers, install from main until a release includes inkling2.

Disclosure and responsibility

Made by OpenUncensored, which runs an API for uncensored models. This model follows instructions it would otherwise
refuse, including harmful ones. You are responsible for how you use it and for complying with the law where you are.
It keeps the Apache 2.0 license of the original.

README history 3 versions

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

  1. 2026-11-13 Update README.md 88a0173 4.2 KB
    Loading...
  2. 2026-11-13 Model card: results on the baked files 5d246c5 3.9 KB
    Loading...
  3. 2026-11-13 Add files using upload-large-folder tool 313fb44 3.8 KB
    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 Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration