Skip to content
Other

Qwen2.5-7B GCAD steering controls

on-the-fly sarcasm and rudeness control

2 min readupdated

A set of ready-made steering banks (activation vectors that modify a model's behavior directly during generation) by RenormalizationTech for Qwen2.5-7B-Instruct. Unlike standard LoRAs or full fine-tunes, the underlying model weights stay untouched — response style is switched dynamically through the inference runtime. Three separate controllers are available: profanity, impolite tone, and sarcasm. Built for LLM steering researchers and developers who need flexible persona control for Qwen2.5-7B bots.

Key features

  • Runs on top of a pinned revision of Qwen/Qwen2.5-7B-Instruct without modifying base weights or using LoRA adapters.
  • Three isolated banks for distinct stylistic traits: profanity, impolite, and sarcasm.
  • Includes a custom Python runtime module with SHA256 manifest integrity checks and a built-in zero-coefficient self-test.
  • Open Apache-2.0 license with no extra usage restrictions.

What you can do

  • Run a local dialogue generator on Qwen2.5-7B-Instruct and dial sarcasm or bluntness up and down by selecting the corresponding bank.
  • Use the prebuilt banks and reports (the repo includes methodology notes and transfer results on a 4B model) for activation steering experiments.

Example

cd controls
python -m runtime.example --snapshot "$LOCAL_BASE_SNAPSHOT" --trait profanity --manifest-sha256 "$BANK_MANIFEST_SHA" --zero-selftest
python -m runtime.example --snapshot "$LOCAL_BASE_SNAPSHOT" --trait profanity --manifest-sha256 "$BANK_MANIFEST_SHA" --prompt "Describe a rainy afternoon in two sentences."
Read moreDetails
Trait expression metrics (baseline to GCAD): PF (b=5) increased from 0 to 31.13 (coherence 85.34 to 82.69); SARC (b=2) rose from 0.06 to 63.78 with coherence dropping from 89.59 to 70.72; Impolite (b=3.5) rose from 1.62 to 77.37 (coherence 82.68 to 75.9).
Requires exact base commit a09a35458c702b33eeacc393d103063234e8bc28 (downloaded separately), Python 3.11, torch 2.6.0+cu124, and transformers 4.52.3.
Controllers were calibrated individually and have not been validated for combined use; sarcasm shows a noticeable coherence drop (-18.88 points).
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 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 2060 · BF16 — from RTX 2060

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

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.

    # RenormalizationTech/qwen2.5-7b-gcad-steering-controls
    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 RenormalizationTech/qwen2.5-7b-gcad-steering-controls --local-dir ~/models/qwen2.5-7b-gcad-steering-controls
    # The model is in ~/models. How to run it is in the model card (link above).
  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

  1. Qwen2.5-7Bfamily root
  2. Qwen2.5-7B-Instruct
  3. qwen2.5-7b-gcad-steering-controls

New posts about versions of Qwen2.5-7B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 1All on Hugging Face
Merges 5All on Hugging Face

About the model

RenormalizationTech/qwen2.5-7b-gcad-steering-controls
Author
RenormalizationTech
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

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

Comments