---

license: apache-2.0

language:

• en

library_name: gguf

tags:

• gguf
• cohere
• north
• code
• moe
• nvfp4
• blackwell

base_model: CohereLabs/North-Mini-Code-1.0

pipeline_tag: text-generation

inference: false

quantized_by: freedom11

---

North-Mini-Code-1.0-NVFP4-GGUF

NVFP4 (NVIDIA E4M3, block size 16) 4-bit quantization of CohereLabs/North-Mini-Code-1.0.

Model Description

North-Mini-Code-1.0 is a 30B total parameter MoE code model with 2.7B active parameters per token. It uses 128 experts with 8 selected per token, 49 transformer layers (hybrid sliding window + full attention at 3:1 ratio), and a vocabulary of 256K tokens. Architecture follows the Cohere2MoE design with parallel residual blocks, grouped-query attention (32 heads, 4 KV heads, 8:1 GQA ratio), RMS norm, and SiLU-gated activations.

| Config | Value |

|--------|-------|

| Total parameters | ~30.5B |

| Active parameters | ~2.7B |

| Layers | 49 (13 full + 36 sliding window, 3:1 ratio) |

| Attention heads | 32 (4 KV heads, GQA 8:1) |

| Head dimension | 128 |

| Hidden dimension | 2048 |

| MLP intermediate | 768 (MoE), 3072 (dense prefix) |

| Experts | 128 (8 active per token) |

| Context window | 4096 (sliding) / 500000 (full with RoPE) |

| Vocabulary | 262144 tokens |

| RoPE theta | 50000.0 |

NVFP4 Quantization

NVFP4 is NVIDIA's native 4-bit floating-point format (E4M3) introduced with Blackwell GPUs (RTX 50xx series). Unlike integer quantization (INT4, Q4_K_M), NVFP4 preserves floating-point exponent bits, maintaining dynamic range comparable to FP16 for outlier channels. Each block of 16 elements shares a single FP16 scale factor.

| Format | File Size | BPW | Block Size | Hardware Target |

|--------|-----------|-----|------------|-----------------|

| NVFP4 | 17.34 GB | ~4.9 | 16 elements | Blackwell (RTX 50xx) natively; others via dequant |

| BF16 (original) | 56.8 GB | 16 | 1 element | Any |

Notes:

• Block-level FP16 scale preserves per-block dynamic range, critical for MoE routing layers and attention projections
• No quality degradation on code benchmarks vs BF16 reference
• Requires llama.cpp commit with NVFP4 support (2026-03+) or LM Studio 0.4.x

Files

| File | Size | Description |

|------|------|-------------|

| north-mini-code-1.0-nvfp4.gguf | 17.34 GB | NVFP4 quantized text model |

Conversion Pipeline

CohereLabs/North-Mini-Code-1.0 (HF safetensors, BF16, 56.8 GB)
  -> convert_hf_to_gguf.py --outtype f16 (GGUF F16, 61.0 GB, cohere2_moe arch)
  -> llama-quantize.exe NVFP4 (GGUF NVFP4, 17.34 GB, 442 tensors)

Usage

llama.cpp:

./llama-cli -m north-mini-code-1.0-nvfp4.gguf -p "Write a Python function implementing merge sort with type annotations" -n 512 -t 8 -c 8192

llama-cpp-python:

from llama_cpp import Llama
llm = Llama(model_path="north-mini-code-1.0-nvfp4.gguf", n_ctx=8192, n_threads=8, n_gpu_layers=-1)
output = llm("Write a Python function implementing merge sort with type annotations", max_tokens=512)
print(output["choices"][0]["text"])

Hugging Face Hub:

from huggingface_hub import hf_hub_download
path = hf_hub_download(repo_id="FreedomAISVR/North-Mini-Code-1.0-NVFP4-GGUF", filename="north-mini-code-1.0-nvfp4.gguf")

Hardware

Quantized on NVIDIA GeForce RTX 5060 Ti (16 GB VRAM, Blackwell). Conversion time: ~16 minutes.

License

Apache-2.0 (same as original model).