automationkasey/simp-master-training

🧠 SIMP Master Training

Fine-tune Qwen2.5-7B-Instruct on 3,003 real SIMP Q&A pairs using QLoRA.

Time: ~20 min on Colab T4 (free)

Output: LoRA adapter pushed to automationkasey/simp-7b-v2

Quick Start

1. Click the "Open in Colab" button above
2. Run Cell 1 (install dependencies)
3. Paste your HF token in Cell 3: HF_TOKEN = "hf_..."
4. Run all cells

The model will learn:

• SIMP protocol architecture (broker, agents, routing)
• Intent routing and agent registration
• Financial Operations (simulated payments, gates, budgets)
• Agent orchestration and security auditing
• System maintenance via ProjectX kernel

After Training

# Clone the adapter
git lfs install
git clone https://huggingface.co/automationkasey/simp-7b-v2

# Merge with base
python -c "
from peft import PeftModel
from transformers import AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained('Qwen/Qwen2.5-7B-Instruct')
model = PeftModel.from_pretrained(model, 'automationkasey/simp-7b-v2')
model = model.merge_and_unload()
model.save_pretrained('./simp-master-merged')
"

# Convert to GGUF for Ollama
# (Use llama.cpp convert script)

Dataset

3,003 pairs from:

• System documentation (73 docs → 2,822 pairs)
• Live broker logs and agent interactions (256 pairs)
• System knowledge files (.goosehints, SIMP_MASTER_CONTEXT.md)

Pipeline

The master training is one part of the SIMP waterfall:

1. Collect system data (187K records)
2. Build Q&A pairs (3,003 examples)
3. Train Master (this notebook — QLoRA on T4)
4. Distill to students (trading-quant, guardian, etc.)
5. Loop every 4 hours