Qwen3.5-0.8B-weight-space-programming-GGUF

Qwen3.5-0.8B-weight-space-programming-GGUF

GGUF copies of a small Qwen3.5 0.8B model whose weights were edited directly to reduce reasoning loops. For people who want to run it locally in several quant sizes.

Made by prithivMLmods on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: This repository holds GGUF files of Qwen3.5-0.8B Weight-Space Programming, a small multimodal model with 752M parameters. The base checkpoint was changed by directly editing 131,072 weights in two MLP layers, with no training or fine-tuning. The edit targeted a loop problem in the model's reasoning.

Who it suits: People who want to run a tiny model locally and study weight editing as an alternative to training.

Industry | Who uses it | How they use it | Example
Research | ML researchers | Study weight editing without gradient descent | Compare the edited model with the original Qwen3.5-0.8B
Software | Developers | Run a tiny model on a low-memory machine | Load the Q4_K_M file in a GGUF program
Education | Students | Learn how quant sizes trade quality for space | Try Q4 and Q8 and compare answers

You can skip it if you need a large model, or if you do not use GGUF programs.

Where to use it:

• llama.cpp and other programs that read GGUF files
• Windows, macOS or Linux, with Python for the download step
• Low-memory machines, using the smaller quants

Use case | Who | What they get back
Math reasoning tests | Researchers | A small model the card says improved on MATH-500 after the edit
Local runs | Developers | Text answers from a 0.8B model on your own machine
Image input | Hobbyists | Vision use through the separate mmproj file

How to set it up:

• Python and pip on your computer
• A program that loads GGUF files, such as llama.cpp
• One quant file, plus the mmproj file if you want images

1. Install the Hugging Face hub package.
2. Download the repository with the hf command.
3. Choose a quant file from the list in the card.
4. Open the card for the run steps.

pip install -U huggingface_hub
hf download prithivMLmods/Qwen3.5-0.8B-weight-space-programming-GGUF

How to use it: Load a quant file in your GGUF program and prompt it. For image input, also load the mmproj-bf16 file. The card marks Q4_K_M, Q4_K_S, Q5_K_M, Q5_K_S and Q6_K as recommended.

Quant | Size and card note
BF16 | 1.52 GB, full weights and highest quality
Q4_K_M | 529 MB, good quality and the default for most uses
Q6_K | 630 MB, high quality, near perfect
Q8_0 | 812 MB, the largest quant, generally unneeded
mmproj-bf16 | 207 MB, the vision projection file

Example: You load the Q4_K_M file and ask a short math question. It returns a worked answer. The card says the original model often got stuck in endless reasoning, which the edit was meant to reduce.

For newcomers and experts: If you are new, download Q4_K_M and ask one math question. It is the card's default size and a small download.

If you know the field, note what the edit touched: 131,072 weights inside two MLP layers, with the vision tower kept. The card reports MATH-500 accuracy of 51.0% versus 5.0% before. It says truncated answers fell to 31.6%, down from 95.0%. Those figures are the original author's, from an H200 run with four generations per problem.

Qwen3.5-0.8B base model
Direct edit of 131,072 weights
Edited checkpoint
GGUF quant files for local runs

Good to know:

• License: Apache 2.0
• Cost to run: none named
• Data it reaches: the files you download to your computer
• Speculative decoding: the card notes MTP is not included

On Hugging Face, 13 November 2026

• Task: image text to text
• Parameters: 752M
• Weights: about 376 MB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 4,807
• Likes: 10
• License: apache-2.0
• Languages: en
• Based on: ML-Intern-lab/Qwen3.5-0.8B-weight-space-programming
• Tags: gguf, activation-steering, llama-cpp, ml-intern, weight-space-programming, image-text-to-text, conversational

Links

• Model: https://huggingface.co/prithivMLmods/Qwen3.5-0.8B-weight-space-programming-GGUF
• Files: https://huggingface.co/prithivMLmods/Qwen3.5-0.8B-weight-space-programming-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download prithivMLmods/Qwen3.5-0.8B-weight-space-programming-GGUF

Then follow the usage in its model card. Nothing in this kit runs by itself.