onocloud/onoai-basalt-2.2-pro

💻 OnoAI Basalt 2.2 Pro (Flagship 32B Sovereign Reasoning & Coding Engine)

The primary workhorse coding and business reasoning engine of the OnoAI ecosystem.
Engineered for autonomous software engineering, full-stack Next.js/React development, OnoCloud ERP ledger logic (ZATCA Phase 2, IFRS), 3D game engines (Three.js/GLSL), and sovereign multi-turn chat.

---

🎯 Key Specifications

• Model Architecture: Advanced Code Transformer (Qwen2.5-Coder-32B-Instruct Core)
• Parameters: ~32.8 Billion Parameters
• Trained Adapters: 16-rank LoRA over all linear projections (q, k, v, o, gate, up, down)
• Training Curriculum: 20,600 Pristine, Deduplicated Samples (0 duplicates, SHA-256 verified)
  • Full-Stack Web: Next.js 16, React 19, Supabase, Tailwind CSS, TypeScript
  • Enterprise ERP & FinTech: ZATCA Phase 2 E-Invoicing, IFRS Fixed Assets, Bank Reconciliation
  • 3D Games & Graphics: Three.js Netcode, Counter-Strike/PUBG Spatial Octrees, Procedural Tracks
  • Autonomous Agents: Gabster-style Agent Marketplace personas (Sales, Cart Recovery, UGC)
• Context Window: 32,768 Tokens
• License: OnoCloud Sovereign Enterprise License (OSRL-1.0) (100% Commercial, Sovereign, Royalty-Free)

---

🚀 Quickstart & Inference

1. Python (Local / RunPod Container via PEFT)

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model_name = "Qwen/Qwen2.5-Coder-32B-Instruct"
adapter_name = "onocloud/onoai-basalt-2.2-pro"

tokenizer = AutoTokenizer.from_pretrained(adapter_name)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)

model = PeftModel.from_pretrained(base_model, adapter_name)
model.eval()

prompt = "<|im_start|>system\nأنت OnoAI Basalt 2.2 Pro، العقل السيادي الموحد للبرمجة والمعمارية ومنظومة سوق الوكلاء.<|im_end|>\n<|im_start|>user\nاكتب دالة TypeScript لحساب ضريبة ZATCA Phase 2 متوافقة مع OnoCloud ERP.<|im_end|>\n<|im_start|>assistant\n"

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=1024, temperature=0.2, top_p=0.95)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

2. Fast vLLM Serving (RunPod / Sovereign Fleet)

vllm serve Qwen/Qwen2.5-Coder-32B-Instruct \
  --enable-lora \
  --lora-modules basalt-pro=onocloud/onoai-basalt-2.2-pro \
  --max-model-len 8192 \
  --gpu-memory-utilization 0.90