Qwen3-VL-8B-Instruct-heretic

Qwen3-VL-8B-Instruct-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 Qwen/Qwen3-VL-8B-Instruct, 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 vision-language knowledge and instruction-following are left largely intact.

Who this is for: developers who want Qwen3-VL's vision-language capabilities - OCR, spatial grounding, GUI/agent use, long-context video understanding - without refusals. Not a capability upgrade over base Qwen3-VL-8B-Instruct - 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 ~8.8B 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 161 of 200, RNG seed 3382198859. Full reproduction assets are in reproduce/.

Files

Safetensors (BF16)

The full-precision weights are in model.safetensors (see the repo file listing for exact sizes).

GGUF quantizations

GGUF quantizations are published for this model (Q4_K_M, Q5_K_M, Q6_K, Q8_0). Pull a specific quant with llama.cpp / ollama.

The GGUFs contain the text backbone (the GGUF format as produced by llama.cpp does not carry the vision encoder); the full vision-language model is available via the Safetensors weights above.

Quickstart

# llama.cpp - defaults to the Q4_K_M quant
llama serve -hf saidutta69/Qwen3-VL-8B-Instruct-heretic:Q4_K_M

# transformers (vision-language)
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
model_name = "saidutta69/Qwen3-VL-8B-Instruct-heretic"
model = Qwen3VLForConditionalGeneration.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained(model_name)
# ... inference code (see Qwen3-VL docs for chat template usage)

Also runnable via Ollama, LM Studio, Jan, vLLM, SGLang.

FAQ

What is Qwen3-VL-8B-Instruct-heretic?

Qwen3-VL-8B-Instruct-heretic is a decensored build of Qwen/Qwen3-VL-8B-Instruct 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 Qwen3-VL-8B-Instruct-heretic actually uncensored?

Yes. On Heretic's 100-prompt refusal harness it drops from 100/100 to 6/100 at KL divergence 0.0313 from the base model. These are measured values, not a claim; reproduction data is in reproduce/.

How do I run Qwen3-VL-8B-Instruct-heretic locally?

Text only: llama serve -hf saidutta69/Qwen3-VL-8B-Instruct-heretic:Q4_K_M (one of 4 GGUF quants). The GGUFs drop the vision tower; use the BF16 safetensors plus AutoProcessor for image input.

Responsible use

Refusal suppression is deliberate and works as intended: this model will comply with requests the base model would refuse, including some it shouldn't. There is no safety filtering layered on top. You are responsible for how you deploy it.

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

License

Inherits the apache-2.0 license from the base model.

Related

• Qwen/Qwen3-VL-8B-Instruct - the base model this checkpoint derives from
• RACER IS OP - Heretic Models - full collection - every heretic model in this collection
• Qwen3-VL-4B-Instruct-heretic - nearest Qwen3-VL sibling
• Qwen3-VL-2B-Instruct-heretic - nearest Qwen3-VL sibling

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.