VibeThinker-3B-heretic

VibeThinker-3B-heretic

<div align="center"> <img src="https://photu.kashyalabanavli.site/racer-is-op.png" alt="RACER IS OP" width="100%"> </div> <br>

A decensored variant of WeiboAI/VibeThinker-3B (fine-tuned from Qwen/Qwen2.5-Coder-3B), produced with Heretic v1.4.0 (directional ablation / "abliteration"). Refusal behavior is suppressed via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the base model's thinking/reasoning gains and instruction-following are left largely intact.

Who this is for: users who want a ~3B reasoning/coding model with an explicit thinking mode (VibeThinker's chain-of-thought behavior) and no refusal guardrails - for CPU and low-VRAM local use, agents, and edge/on-device reasoning. Not a capability upgrade over base VibeThinker-3B - same model, refusal guardrails removed.

<!-- racer-gpu-matrix -->

Runs on your gaming PC

Full GGUF ladder included — pick the quant that fits your card:

Weights only, at this model's 3.1B native size; add ~1 GB for context. OOM? Drop one quant level. Headroom to spare? Go one up.

Why abliteration instead of fine-tuning

Fine-tuning a "helpful" persona on top of RLHF'd refusals fights the base model's training and tends to degrade coherence. Abliteration instead finds and edits the specific weight directions responsible for refusal, leaving the rest of the network (and its capabilities) untouched. See the Heretic repo and the original abliteration writeup for the mechanism.

Performance

Reproduced with Heretic 1.4.0 (directional ablation), trial 92 of 200, RNG seed 4063399784. Full reproduction assets are in reproduce/.

Files

GGUF quantizations

Full quantization set (14 quants + F16) produced with llama.cpp.

Qwen2.5-Coder architecture with VibeThinker thinking mode — loads natively in llama.cpp / Ollama / LM Studio / Jan.

Run llama serve -hf saidutta69/VibeThinker-3B-heretic to pull the default quant.

Quickstart

llama.cpp

# download the LTS build: https://github.com/ggerganov/llama.cpp/releases
llama-cli -m VibeThinker-3B-heretic-Q4_K_M.gguf -p "User: What is the fourth dimension?\nAssistant:" -n 512

Ollama

ollama run hf.co/saidutta69/VibeThinker-3B-heretic:Q4_K_M

LM Studio

1. Open LM Studio and click the search icon to open the Model Search panel.
2. Type "VibeThinker-3B-heretic" and click the download button marked GGUF.
3. Pick your quant, load the model, and start chatting.

FAQ

What is VibeThinker-3B-heretic?

VibeThinker-3B-heretic is a decensored build of WeiboAI/VibeThinker-3B produced with Heretic 1.4.0 (directional ablation). It is the same weights with the refusal direction removed, not a new fine-tune, so its capabilities track the base model.

Is VibeThinker-3B-heretic actually uncensored?

Yes. On Heretic's 100-prompt refusal harness it drops from 57/100 to 22/100 at KL divergence 0.0444 from the base model. These are measured values, not a claim; reproduction data is in reproduce/. This edit is conservative: only 35 of 57 refusals cleared, so read KL and refusal count together.

How do I run VibeThinker-3B-heretic locally?

llama serve -hf saidutta69/VibeThinker-3B-heretic:Q4_K_M pulls the default quant; the repo ships 15 GGUF quants (see Files).

Responsible use

This model has its refusal guardrails removed. Use it lawfully and ethically, and be aware that it may generate content the base model would refuse.

Made with ❤️ by RACER IS OP (saidutta69) — follow for more uncensored models

Related

• WeiboAI/VibeThinker-3B - the base model this checkpoint derives from
• RACER IS OP - Heretic Models - full collection - every heretic model in this collection

Abliteration hardware

Recorded in reproduce/reproduce.json:

• GPU: NVIDIA L4 (22.03 GB VRAM)
• CPU: Intel Xeon Processor (SapphireRapids)
• OS: Linux-6.7.10-060710-generic-x86_64-with-glibc2.39
• CUDA / driver: CUDA 12.8, driver 580.126.20
• Software: PyTorch 2.11.0+cu128, Heretic 1.4.0

Heretic notes that KL-divergence and refusal numbers can vary slightly with hardware and library versions.