Skip to content
Computer vision

Inkling-2-Mini Uncensored LoRA

removing restrictions from a compact multimodal model

2 min readupdated

This model is a decensored version of the thinkingmachines/Inkling-2-Mini multimodal base model, obtained using the Arbitrary-Rank Ablation (ARA) method. It accepts text and image inputs and generates a text response, retaining the original architectural features while significantly lowering the refusal rate on requests. This new release suits developers and researchers who need flexible model behavior without rigid built-in filters.

Key features

  • Minimal refusals: the refusal rate drops from 100/100 in the original to 3/100.
  • ARA method: targeted layer modification (layers 1 to 16) while preserving useful behavior.
  • Base architecture: retains the original's multimodal capabilities and long context support.
  • Minimal divergence: the KL divergence metric is kept low at 0.0883.

What you can do

  • Pipeline integration: use as an adapter for local applications with increased tolerance for complex prompts.
  • Safety research: study restriction removal methods and the impact of ARA parameters on transformer behavior.
  • Agent testing: run autonomous agents for programming and image analysis tasks without false filter triggers.

Example

Usage examples and ARA configuration parameters to reproduce the model are available in the repository under the reproduce directory.

Read moreDetails
Metrics: Refusals 3/100 vs 100/100 for the original.
KL divergence: 0.0883.
ARA parameters: start_layer_index = 1, end_layer_index = 16, loss_mode = relative.
Hardware requirements match the base Inkling 2 Mini model (32GB+ VRAM or supported local devices).
Discuss Telegram 1

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • U8weights 26 GB · ready files30 GB
  • FP8 / INT8recommendedweights 25 GB28 GB
  • 4-bitweights 14 GB16 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 5090 · U8 — from RTX 5090

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model Inkling-2-Mini.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # electroglyph/Inkling-2-Mini-uncensored-LoRA
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download thinkingmachines/Inkling-2-Mini --local-dir ~/models/Inkling-2-Mini
    hf download electroglyph/Inkling-2-Mini-uncensored-LoRA --local-dir ~/models/Inkling-2-Mini-uncensored-LoRA
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Inkling-2-Mini
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

thinkingmachines/Inkling-2-Mini

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of Inkling-2-Mini (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

electroglyph/Inkling-2-Mini-uncensored-LoRA
Author
electroglyph
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

01

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments