DEV.co
Open-Source LLM · bartowski

North-Mini-Code-1.0-GGUF

North-Mini-Code-1.0-GGUF is a quantized version of Cohere's North-Mini-Code model, optimized for local inference via llama.cpp. It is a code-capable conversational model available in 27 quantization variants ranging from 8.51 GB to 61 GB, allowing trade-offs between quality and resource consumption. The model is ungated, Apache 2.0 licensed, and can run on consumer hardware with modest VRAM depending on quantization choice.

Source: HuggingFace — huggingface.co/bartowski/North-Mini-Code-1.0-GGUF
Unknown
Parameters
apache-2.0
License (OSI-approved)
Unknown
Context (tokens)
39.4k
Downloads (30d)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

FieldValue
Developerbartowski
ParametersUnknown
Context windowUnknown
Licenseapache-2.0 — OSI-approved
Modality / tasktext-generation
Gated on HuggingFaceNo
Downloads39.4k
Likes2
Last updated2026-06-14
Sourcebartowski/North-Mini-Code-1.0-GGUF

What North-Mini-Code-1.0-GGUF is

This is a GGUF-format quantization distribution of CohereLabs/North-Mini-Code-1.0, quantized using llama.cpp release b9630 with imatrix calibration. The repository provides 27 quantization variants (bf16, Q8_0 through Q2_K, IQ variants) optimized for different CPU/GPU inference trade-offs. Quantization quality ranges from bf16 (uncompressed) down to IQ2_XXS (extreme compression). The model uses a custom chat prompt format with system/user/chatbot token structure and includes tool-calling scaffolding. Parameter count and context length are not documented.

Quickstart

Run North-Mini-Code-1.0-GGUF locally

Load the open weights with 🤗 Transformers and generate — the same model, self-hosted.

quickstart.pypython
from transformers import pipelinepipe = pipeline("text-generation", model="bartowski/North-Mini-Code-1.0-GGUF")out = pipe("Explain retrieval-augmented generation in one sentence.",           max_new_tokens=128)print(out[0]["generated_text"])

Swap in vLLM or Ollama for production-grade serving. DEV.co can stand up the inference stack.

Deployment

How you'd run it

A typical self-hosted path — open weights, an inference server, your application.

DEV.co builds each layer — from GPU infrastructure to the application.

Best use cases

Local code assistance with minimal hardware

Use IQ4_XS or Q4_K_M quantizations (16–18 GB) on consumer GPUs or high-end CPUs for real-time code completion, debugging, and refactoring without cloud dependencies or latency.

Embedded agent systems in production

Deploy as a local agent backbone in resource-constrained environments (IoT, edge devices) using Q3_K_L or lower quantizations; tool-calling format supports function binding.

Privacy-sensitive coding workflows

Self-host in air-gapped or on-premise environments where code and prompts must never leave the network; Apache 2.0 license permits commercial deployment.

Running & fine-tuning it

Highly dependent on quantization choice. ESTIMATE: Q4_K_M (~18.74 GB, recommended for general use) requires 24–32 GB system memory for inference + context. Q6_K (~26.4 GB) needs 32–40 GB. Q2_K (~11 GB) can run on 16 GB systems with aggressive context limits. GPU acceleration (CUDA, Metal) is supported via llama.cpp; CPU-only inference is feasible but slower. Context length not documented; assume typical 4k–8k unless verified.

Model card does not discuss fine-tuning, LoRA, or QLoRA. Quantized GGUF format is designed for inference only. To fine-tune, obtain the original CohereLabs/North-Mini-Code-1.0 model in full precision, use standard HF Transformers + PEFT for LoRA. Quantization may impede gradient flow; requires validation on target hardware.

When to avoid it — and what to weigh

  • Extreme latency sensitivity with sub-100ms SLA — Quantized model inference via llama.cpp on CPU will not meet strict real-time requirements; measure on target hardware first.
  • Unsupported or legacy code languages — No model card details on supported programming languages, training data composition, or language-specific accuracy; validate against your tech stack before adoption.
  • You need model fine-tuning on proprietary data — No LoRA, QLoRA, or fine-tuning guidance provided. Quantization may complicate retraining workflows; investigate compatibility with your fine-tuning stack.
  • Benchmark-driven selection without validation — No published benchmarks (MBXP, HumanEval, or similar) are provided; claimed performance is unverified. Evaluate output quality on representative code samples before production use.

License & commercial use

Apache License 2.0 (OSI-approved permissive license). Permits commercial use, modification, and distribution with attribution and indemnification notice. No patent grants or trademark restrictions in the excerpt provided.

Apache 2.0 is a permissive OSI license that explicitly permits commercial use. However, verify that the original base model (CohereLabs/North-Mini-Code-1.0) carries no additional restrictions. The quantization distribution itself can be used commercially, but ensure compliance with any Cohere terms of service or trademark policy for the 'North-Mini-Code' naming if deploying as a branded service. Obtain legal review if bundling with proprietary code.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

SignalAssessment
MaintenanceActive
DocumentationLimited
License clarityClear
Deployment complexityLow
DEV.co fitGood
Assessment confidenceMedium
Security considerations

Quantized GGUF format runs locally (no cloud transmit); this mitigates data exfiltration risk if infrastructure is isolated. GGUF format is not a security guarantee—the model weights are unencrypted on disk. Potential concerns: (1) Model may inherit biases or unsafe outputs from training data; (2) Tool-calling interface could be exploited if bound to untrusted functions; (3) No safety audit or red-teaming results published. Validate model behavior (jailbreak resistance, instruction adherence) on sensitive use cases before production.

Alternatives to consider

mistral-7b-instruct-v0.2 (Mistral AI; GGUF available)

Smaller, widely benchmarked, strong code capability, extensive community support and serving options; fewer quantization variants but lower hardware floor.

codellama-34b-instruct (Meta; GGUF available)

Purpose-built for code, trained on large code corpora, better documented instruction-following; larger (34B) but available in low quantizations; stronger benchmark presence.

granite-code-instruct-8k (IBM; Apache 2.0, GGUF available)

Smaller footprint (8B params), Apache-licensed, designed for code tasks, enterprise-friendly; less download activity but comparable license clarity and local-first design.

Software development agency

Ship North-Mini-Code-1.0-GGUF with senior software developers

Download the right quantization for your hardware, validate on representative code samples, and integrate via llama.cpp or your preferred serving stack. Consult the original base model docs and benchmark against alternatives before production use.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

Related on DEV.co

Explore the category and the services that help you build with it.

North-Mini-Code-1.0-GGUF FAQ

Can I use this commercially?
Yes. The GGUF quantizations are Apache 2.0 licensed, which permits commercial use with attribution. However, verify the original CohereLabs/North-Mini-Code-1.0 model has no additional commercial restrictions. Consult legal counsel if rebranding or bundling with proprietary code.
Which quantization should I choose for my hardware?
Q4_K_M (18.74 GB) is labeled 'recommended' for general use and strikes a balance between quality and size. For constrained hardware (<20 GB VRAM/RAM), use IQ4_XS (16.58 GB) or Q4_K_S (18.05 GB). For maximum quality with sufficient memory (>32 GB), use Q6_K (26.4 GB). Test on your target machine; VRAM/performance is not documented.
What is the context length?
Not documented in the model card. Check the original CohereLabs/North-Mini-Code-1.0 repository or test empirically. Assume default transformer sizes (4k–8k) unless verified.
Can I fine-tune this quantized model?
GGUF is an inference-only format. To fine-tune, obtain the original CohereLabs/North-Mini-Code-1.0 in full precision, then use Hugging Face Transformers + PEFT (LoRA). Quantization complicates gradient flow; validate compatibility with your fine-tuning stack before committing.

Software development & web development with DEV.co

DEV.co is a software development agency delivering custom software development services to companies building on open source. Our software developers and web developers design, integrate, and ship production systems — spanning web development, APIs, AI, data, and cloud. If North-Mini-Code-1.0-GGUF is part of your open-source llms roadmap, our team can implement, customize, migrate, and maintain it.

Ready to deploy a local code assistant?

Download the right quantization for your hardware, validate on representative code samples, and integrate via llama.cpp or your preferred serving stack. Consult the original base model docs and benchmark against alternatives before production use.