North-ML1/Ink-1-GGUF overview
Ink 1 banner.png Ink 1 GGUF GGUF container build of North ML1/Ink 1 https://huggingface.co/North ML1/Ink 1 ~725M params . Files | File | Size | Contents | | | …
Runs locally from ~789.3 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| Ink-1-8bit.gguf | GGUF | GGUF | 789.3 MB | Download |
Model Details
Model README
---
license: apache-2.0
tags:
- gguf
- north-ml
- flare
- deltanet
thumbnail: banner.png
---
Ink 1 GGUF
GGUF container build of North-ML1/Ink-1 (~725M params).
Files
| File | Size | Contents |
|---|---|---|
| Ink-1-8bit.gguf | 828 MB | All 1,368 tensors - identical to model.safetensors |
Important: not llama.cpp-compatible
This is a container-format build. The flare-1 architecture (gated DeltaNet
hybrid + AgentRegisters + Fiber trunk + router head) is not supported by
llama.cpp / Ollama / LM Studio. The GGUF carries the exact same tensors as the
safetensors artifact for tooling that prefers the format:
- packed 8-bit quantized matrices kept bit-identical (U32 stored as I32 bitcast)
- BF16 support tensors losslessly upcast to F32
- all flare metadata (
flare.*keys) preserved from the safetensors header
To run the model, use the safetensors artifact with the Kestrel MLX runtime.
Part of North ML - open weights, honest numbers.
Run North-ML1/Ink-1-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models