Skip to content
Computer vision

Qwen3.5-0.8B-NKC

modified weights for logic and physics

2 min readupdated

Qwen3.5-0.8B-NKC builds on the compact Qwen/Qwen3.5-0.8B language model (873.4 million parameters) using the Neural Knowledge Compiler (NKC) method, which compiles skills from other models directly into existing weights without adding new parameters. By modifying just 0.38% of the weights, the developers significantly boosted its mathematical thinking mode and gave it a basic grasp of scene depth and motion physics. The model loads with the standard transformers library and requires no custom inference code, making it useful for developers and researchers who need a lightweight model with stronger reasoning capabilities.

Key features

  • Zero parameter overhead: All upgrades are written directly into the base model's existing tensors.
  • Major math gains: In thinking mode, scores jumped from 6% to 53% on MATH-500 and from 16% to 76% on GSM8K.
  • Added depth estimation: Integrating data from Depth Anything V2 improved its ability to judge object distance.
  • Physics prediction: Trained on JEPA-WM data to predict whether a push will move an object.

What you can do

  • Solve math and logic problems with high accuracy using thinking mode (enable_thinking=True).
  • Test spatial perception and depth estimation on images using the built-in vision layers.
  • Drop the model into local chatbots and text-and-image pipelines via a standard transformers setup.

Example

from transformers import AutoModelForImageTextToText, AutoProcessor

model = AutoModelForImageTextToText.from_pretrained("MichielH/Qwen3.5-0.8B-NKC", dtype="bfloat16", device_map="auto")
processor = AutoProcessor.from_pretrained("MichielH/Qwen3.5-0.8B-NKC")

messages = [{"role": "user", "content": [{"type": "text", "text": "What is 17 * 23? Please reason step by step, and put your final answer within \\boxed{}."}]}]
inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", enable_thinking=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=4096, do_sample=True, temperature=0.6, top_p=0.95, top_k=20)
print(processor.decode(out[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))
Read moreDetails
• Base model: Qwen/Qwen3.5-0.8B (873.4M parameters)
• Modified weights: ~3.28M (0.38% of total)
• Benchmarks: MATH-500 (with thinking) jumped from 6% to 53%; GSM8K rose from 16% to 76%
• Limitations: Math performance may dip slightly in default mode without thinking enabled
Discuss Telegram 0 12

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.6 GB · ready files2.3 GB
  • FP8 / INT8recommendedweights 0.8 GB1.4 GB
  • 4-bitweights 0.5 GB1.0 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.

    # MichielH/Qwen3.5-0.8B-NKC
    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 MichielH/Qwen3.5-0.8B-NKC --local-dir ~/models/Qwen3.5-0.8B-NKC
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Qwen3.5-0.8B-NKC
  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. Qwen3.5-0.8B-Basefamily root
  2. Qwen3.5-0.8B
  3. Qwen3.5-0.8B-NKC

New posts about versions of Qwen3.5-0.8B-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 9All on Hugging Face
Merges 1All on Hugging Face

About the model

MichielH/Qwen3.5-0.8B-NKC
Author
MichielH
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

120

Ask AI about this model

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

Comments