arthu1/wind-arc-1-6-beta

Wind Arc 1.6

by North.ai

"Is GPT your personal assistant? Well, look at ours."

Wind Arc is a custom architecture language model built for coding, Christian guidance, and everyday assistance. Trained on an RTX 5090 for $1 by an 11 year old.

---

What makes it different

Feature	Details
Custom Architecture	YaRN RoPE (32k context) + MoE FFN (4 routed experts + 1 shared) + Hybrid Attention
Runs locally	No cloud, no subscription, no privacy concerns
Open source	Apache 2.0 — download, run, modify, build on top
Built by North.ai	AI startup focused on powerful AI for small hardware
Trained by an 11 year old	On a $1 RTX 5090 GPU budget

---

Architecture

Wind Arc 1.6 modifies Qwen3-1.7B with:

• YaRN RoPE — extends context from 8k → 32k tokens
• Mixture of Experts FFN — 4 routed experts + 1 shared expert per layer (replaces standard MLP)
• Hybrid Attention — full attention every 4th layer, sliding window (2048) otherwise
• QK-Norm — training stability (Qwen3 style)

Total parameters: ~3.6B (MoE expanded)

---

Good at

• Agentic coding — Python, JavaScript, TypeScript, AI/ML with explanations
• Christian guidance — Bible questions, theology, Christian living, prayer
• Learning — teaches step by step, not just dumps answers
• General chat — questions, research, writing

---

Quick Start

Python (transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "arthu1/wind-arc-1-6",
    torch_dtype=torch.bfloat16,
    trust_remote_code=True,
)
tok = AutoTokenizer.from_pretrained("arthu1/wind-arc-1-6")

prompt = "<|im_start|>system\nYou are Wind Arc, a helpful AI by North.ai.<|im_end|>\n<|im_start|>user\nWho are you?<|im_end|>\n<|im_start|>assistant\n"
inputs = tok(prompt, return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=200, temperature=0.7, do_sample=True)
print(tok.decode(out[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

MLX (Apple Silicon)

pip install mlx-lm
mlx_lm.generate \
  --model arthu1/wind-arc-1-6 \
  --prompt "<|im_start|>user\nWrite a Python binary search<|im_end|>\n<|im_start|>assistant\n" \
  --max-tokens 300

---

System Prompt

For best results use this system prompt:

You are Wind Arc, a friendly and knowledgeable AI assistant made by North.ai. 
You are an expert coder who explains code clearly. 
You approach life from a Christian worldview.

---

Example Outputs

Coding:

User: Write a Python binary search with explanation
Wind Arc: def binary_search(arr, target):
    left, right = 0, len(arr) - 1
    while left <= right:
        mid = (left + right) // 2
        if arr[mid] == target: return mid
        elif arr[mid] < target: left = mid + 1
        else: right = mid - 1
    return -1
# O(log n) — halves the search space each step.
# For 1M items, at most 20 comparisons.

Christian:

User: What is the Gospel?
Wind Arc: The Gospel means 'good news' — Jesus Christ died for human 
sin and rose from the dead. Through faith in him, anyone receives 
forgiveness and eternal life. John 3:16: God so loved the world that 
he gave his only Son, that whoever believes in him shall not perish 
but have eternal life.

---

---

Roadmap

• Wind Arc 1.6 SFT (supervised fine-tuning for better instruction following)
• GGUF conversion for LM Studio
• Wind Arc 2.0 (more compute, better data)
• Neurotype platform launch
• North Star architecture (fully custom, no base model)

---

About North.ai

North.ai is an AI startup with one mission: powerful AI that runs on small hardware.

Our platform, Neurotype, lets anyone use, train, and deploy AI models without needing a huge cloud budget.

Wind Arc is our flagship model — built to run on your machine, not ours.

• 🌐 Website: https://north-ml.vercel.app
• 🤗 HuggingFace: arthu1

---

License

Apache 2.0 — use it, modify it, build on it. Just keep the North.ai attribution.

---

Wind Arc 1.6 — trained on RTX 5090, built on M4 Mac Mini, by an 11 year old.