North-Mini-Code-1.0 GGUF

North-Mini-Code-1.0 GGUF

GGUF copies of the North-Mini-Code-1.0 coding model, compressed with ByteShape's ShapeLearn for GPU use in llama.cpp. Four sizes from 12.1 GB to 21.5 GB, for developers with a single GPU.

Made by byteshape 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: North-Mini-Code-1.0 GGUF is a set of quantized copies of CohereLabs' North-Mini-Code-1.0 model, a 30.5B text-generation model for code. ByteShape made them with ShapeLearn, which picks a datatype for each tensor to keep quality high even at low bit counts. The files are tuned for GPU inference.

Who it suits: It suits developers who run a coding model locally in llama.cpp and want a size that fits their GPU. The card gives a rule for choosing between the four files.

Title: Who uses it
Industry | Who uses it | How they use it | Example
Software | Application developers | Run a code model on their own GPU | Start llama-server with one of the four tags
AI tooling | Local-model users | Pick the size that suits their graphics card | Choose the smallest file that fits memory
Research | Engineers | Compare quality against speed across sizes | Read the card's quality and tokens-per-second chart

People without a GPU can skip it, because these files are optimized for GPU inference.

Where to use it:

• llama.cpp, through llama-server
• A web interface on localhost once the server runs
• Windows, macOS or Linux
• GPUs such as the RTX 4090, 4080, 5060Ti and 5090, which the card benchmarks

Title: Top jobs
Use case | Who | What they get back
Local coding help | Developers | Code answers from a model on their own GPU
Size selection | Local-model users | A file whose size and speed match their hardware
Benchmark reading | Researchers | Quality scores set against the BF16 baseline

How to set it up:

• Python on your computer, for the download
• The huggingface_hub package
• llama.cpp installed
• A GPU with enough memory for the file you choose

1. Install the Hugging Face hub package.
2. Download the files, or skip this and let llama.cpp fetch one on first run.
3. Pick a size from the table below.
4. Start llama-server with that file's tag.

pip install -U huggingface_hub
hf download byteshape/North-Mini-Code-1.0-GGUF

Title: The four files
Model ID | Bits per weight | File size
GPU-1 | 3.17 | 12.1 GB
GPU-2 | 4.14 | 15.8 GB
GPU-3 | 4.27 | 16.3 GB
GPU-4 | 5.64 | 21.5 GB

How to use it: Copy a model tag and give it to llama-server. llama.cpp downloads the model on the first run. The server then offers a web interface on localhost.

llama-server -hf <MODEL_TAG>

These are the four tags, in the order of the table above.

byteshape/North-Mini-Code-1.0-GGUF:North-Mini-Code-1.0-IQ3_S-3.17bpw
byteshape/North-Mini-Code-1.0-GGUF:North-Mini-Code-1.0-IQ4_XS-4.14bpw
byteshape/North-Mini-Code-1.0-GGUF:North-Mini-Code-1.0-IQ4_XS-4.27bpw
byteshape/North-Mini-Code-1.0-GGUF:North-Mini-Code-1.0-IQ4_XS-5.64bpw

Example: You choose GPU-2 for a 16 GB card, run llama-server with its tag and wait for the download. You open the web interface and ask for a function in your language of choice. The model replies with code in the chat window.

For newcomers and experts: If you are new, pick the largest file that fits your GPU memory, since bigger files keep more quality. A first try is the web interface and one small coding question.

For experts, the card's rule is simple: take the best quality at the speed you need, or the quickest file that still reaches the quality you require. Quality uses LiveCodeBench V6 results, plus MultiPL-E HumanEval results, all with thinking on and divided by the BF16 baseline. The tags such as IQ4_XS are only labels, and they do not follow llama.cpp's usual profiles.

Pick a file by size and speed
Copy its model tag
Run llama-server with the tag
Chat in the local web interface

Good to know:

• Labels: names such as IQ4_XS mark the main method and average bit length, so several files can share one.
• GPU focus: the files are optimized for GPU inference, and the benchmarks run on GPUs.
• License: Apache 2.0.
• Cost to run: none named, and you need a GPU with enough memory.
• Data it reaches: it downloads files from Hugging Face, and the server runs on your own machine.
• Not for: CPU-only machines.

Title: File size by model ID
GPU-1: 12.1 GB
GPU-2: 15.8 GB
GPU-3: 16.3 GB
GPU-4: 21.5 GB

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 30.5B
• Weights: about 15.2 GB at 4-bit (estimate)
• Formats: GGUF
• Downloads in the last 30 days: 269
• Likes: 10
• License: apache-2.0
• Based on: CohereLabs/North-Mini-Code-1.0
• Tags: transformers, gguf, agent, cohere, byteshape, text-generation, conversational

Links

• Model: https://huggingface.co/byteshape/North-Mini-Code-1.0-GGUF
• Files: https://huggingface.co/byteshape/North-Mini-Code-1.0-GGUF/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download byteshape/North-Mini-Code-1.0-GGUF

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