HIGH NOON TRADER

PHASE 2 — LOCAL AI SETUP (MacBook Pro M5 • 32GB)

·

by

Happy Harry Huckleberry
Skip to main content
< All Topics
Print
Updated

🎯 Goal

By the end of Phase 2 you will have:

  • a working local AI setup
  • tested 7B, 14B, and 20B models
  • identified your primary model (14B)
  • established a dual-model workflow (14B + 20B)
  • a functioning AI that can analyze your trading data

🧠 OVERALL STRATEGY

We use two tools, with clear roles:

🧪 LM Studio — Primary (Start Here)

  • model testing
  • performance validation
  • daily interaction (initially)

⚙️ Ollama — Secondary (Production Engine)

  • background model serving
  • future automation + dashboard integration

🧩 STEP 1 — INSTALL LM STUDIO (START HERE)

  1. Download and install LM Studio
  2. Open → go to Model Search

🎯 Download THESE types of models

🟢 7B (fast utility)

  • Mistral 7B Instruct
  • Phi-3 Mini

🟡 14B (PRIMARY MODEL — most important)

  • Qwen 2.5 14B Instruct
  • DeepSeek 14B Distill

🔵 20B (DEEP ANALYSIS MODEL)

  • Qwen 20B (Instruct/Distill)
  • DeepSeek R1 Distill 20B

⚠️ Quantization (critical)

Choose:

  • ✅ Q4_K_M or Q5_K_M

Avoid (for now):

  • ❌ Q8 (too heavy, slower)

🧪 STEP 2 — TEST PERFORMANCE (7B → 14B → 20B)

Start with 7B

Explain a simple momentum trading strategy

✔ Confirm:

  • fast response
  • smooth interaction

Move to 14B (your main model)

Compare momentum vs mean reversion in high volatility markets

✔ Evaluate:

  • reasoning quality
  • structure
  • speed

Now test 20B (important)

Analyze how these strategies behave differently across bull, bear, and high volatility regimes and explain why

✔ Observe:

  • deeper reasoning
  • more nuance
  • slower response

🧠 STEP 3 — TEST ON YOUR REAL DATA

Go to your folder:

AI_TRADING_SYSTEM/

Take a small sample from:

  • strategy file
  • backtest summary

Prompt:

Given this strategy data, analyze strengths, weaknesses, and performance across market regimes

👉 This is your first real:

AI trading analysis using your system

⚖️ STEP 4 — ESTABLISH YOUR WORKFLOW (CRITICAL STEP)

This is the most important part of Phase 2.

🟡 Step 1 — Use 14B FIRST

Compare these strategies and identify strengths and weaknesses

👉 Fast, iterative thinking

🔵 Step 2 — Use 20B SECOND

Critique this analysis. What risks or blind spots were missed?

👉 Deep reasoning + validation

🔁 Step 3 — Iterate if needed

Go back to 14B with refined questions.

🧠 YOUR WORKFLOW BECOMES:

14B → explore
20B → validate
14B → refine

⚙️ STEP 5 — INSTALL OLLAMA (AFTER TESTING)

Now install Ollama

Run your first model:

ollama run mistral

or

ollama run qwen:14b

Test:

What are the risks of a high Sharpe strategy with large drawdowns?

👉 This confirms:

  • backend is working
  • models run outside GUI

🧠 STEP 6 — UNDERSTAND YOUR SYSTEM NOW

At this point:

✔ You HAVE:

  • local AI running
  • 7B / 14B / 20B tested
  • real trading analysis working
  • defined workflow

❌ You DO NOT HAVE YET:

  • RAG (auto-reading your folders)
  • automation
  • dashboard UI

⚠️ COMMON MISTAKES (AVOID THESE)

❌ Using 20B for everything
❌ Downloading too many models
❌ Feeding huge datasets directly
❌ Expecting perfect answers immediately

🏁 PHASE 2 SUCCESS CRITERIA

You are done when:

✔ You ran 7B, 14B, and 20B
✔ You identified 14B as your main model
✔ You used 20B for deeper validation
✔ You tested your real trading data
✔ You ran a model in Ollama

💡 FINAL TAKEAWAY

You now have a multi-level AI analyst system:

  • fast thinking (7B)
  • strong reasoning (14B)
  • deep validation (20B)