Skip to content
Computer vision

Aurum 4B

an uncensored Qwen3.5-4B variant with zero refusals

2 min read

Aurum 4B is the official multimodal Qwen/Qwen3.5-4B model (4.66 billion parameters) stripped of refusal behavior while keeping all remaining weights intact, including the original vision tower and MTP (multi-token-prediction) head. Built for everyday chat and tool calling on GPUs with 8 GB VRAM, it handles complex or sensitive prompts without refusal while maintaining high accuracy on instruction-following and function-calling benchmarks compared to the base model.

Key features

  • Unrestricted responses: 0 refusals out of 100 test harmful prompts, with a 0% false-positive rate on benign requests.
  • Hardware optimization: engineered to run on 8 GB VRAM GPUs, with GGUF quantizations supported for llama.cpp.
  • Intact multimodality: retains the original vision projector and tokenizer from the base Qwen3.5 model.
  • Stronger benchmarks: outperforms its predecessors on IFEval and BFCL (Function Calling Leaderboard).

What you can do

  • Plug into local chatbots: deploy the model in a local app or bot for unrestricted text and code generation without extra censorship.
  • Test tool calling: power projects that need reliable function calling on modest hardware.
  • Run multimodal workflows: process both images and text on the Qwen3.5 architecture with extended reasoning capabilities.
  • Run via llama.cpp: use ready-made GGUF quantizations for fast inference on consumer PCs.

Example

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained("2Beyes/Aurum_4b-safetensors", dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained("Qwen/Qwen3.5-4B")
Read moreDetails
Benchmarks: IFEval strict (78.6), BFCL non-live average (85.9), GPQA Diamond (49.5).
Requirements: GPU with at least 8 GB VRAM.
Limitations: the model may occasionally trigger a tool when none is needed (BFCL irrelevance 80.8) and has only been tested in English.
License: Apache License 2.0.
Discuss Telegram 0 24

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 8.7 GB · ready files10 GB
  • FP8 / INT8recommendedweights 4.3 GB5.3 GB
  • 4-bitweights 2.4 GB3.2 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 2080 Ti

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.

    # 2Beyes/Aurum_4b-safetensors
    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 2Beyes/Aurum_4b-safetensors --local-dir ~/models/Aurum_4b-safetensors
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Aurum_4b-safetensors
  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. Qwen3.5-4B-Basefamily root
  2. Qwen3.5-4B
  3. Aurum_4b-safetensors

New posts about versions of Qwen3.5-4B-Base (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

2Beyes/Aurum_4b-safetensors
Author
2Beyes
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

240

Ask AI about this model

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

Comments