HIGH NOON TRADER

🟡 DAY 2 — LOCAL AI SETUP (OLLAMA + MODEL STACK)

·

by

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

🎯 Goal

By the end of Day 2 you will have:

  • Ollama running locally on your Mac
  • 7B / 14B / 20B models installed
  • verified model performance differences
  • first basic “strategy analysis” prompts working on your system

This is where your system becomes an AI research engine, not just folders.

💻 STEP 1 — INSTALL OLLAMA

Install (Mac)

Run:

curl -fsSL https://ollama.com/install.sh | sh

Verify install:

ollama --version

You should see version output.

🧠 STEP 2 — DOWNLOAD YOUR MODEL STACK

We install your 3-layer reasoning system:

🟢 7B MODEL (fast utility layer)

ollama pull llama3.1:8b

(acts as your “light analysis / quick response” model)

🟡 14B MODEL (main analysis engine)

ollama pull qwen2.5:14b

This becomes your default:

  • strategy analysis
  • comparisons
  • reasoning over backtests

🔴 20B MODEL (deep reasoning layer)

ollama pull deepseek-r1:20b

Used for:

  • edge-case reasoning
  • conflicting strategy evaluation
  • deep failure analysis

⚙️ STEP 3 — TEST EACH MODEL

Test 7B

ollama run llama3.1:8b

Prompt:

Summarize a momentum trading strategy in 5 bullet points.

Test 14B

ollama run qwen2.5:14b

Prompt:

Compare momentum vs mean reversion in volatile markets.

Test 20B

ollama run deepseek-r1:20b

Prompt:

Why do some strategies perform well in backtests but fail in live conditions? Use structured reasoning.

🧠 STEP 4 — DEFINE MODEL ROLES (CRITICAL)

Create file:

touch 08_MODEL_NOTES/model_roles.md

Paste:

7B MODEL:
- fast responses
- tagging data
- simple summaries

14B MODEL:
- main analysis engine
- strategy comparisons
- regime reasoning

20B MODEL:
- deep reasoning
- contradiction analysis
- edge-case evaluation

🧪 STEP 5 — FIRST “TRADING SYSTEM TEST PROMPT”

Now test your system like it’s already Phase 4:

Run:

ollama run qwen2.5:14b

Then paste:

You are analyzing historical trading strategies.
Compare momentum vs mean reversion in high volatility regimes.
Structure your answer with: Summary, Performance, Risk, and Conclusion.

🧠 STEP 6 — CONNECT TO YOUR FUTURE SYSTEM (IMPORTANT)

Right now you are NOT connecting to RAG yet.

But you are verifying:

✔ model behavior
✔ reasoning quality
✔ response structure

This ensures Phase 4 will work cleanly later.

⚠️ COMMON ISSUES (EXPECTED)

If model is slow:

  • normal on first run (loading weights into memory)

If RAM spikes:

  • expected on 14B/20B
  • keep only one model running at a time

If 20B is slow:

  • that is OK → it is your “deep analysis only” layer

🧱 WHAT YOU NOW HAVE (END OF DAY 2)

You now have:

✔ local AI runtime (Ollama)
✔ 3-tier model system (7B / 14B / 20B)
✔ verified reasoning behavior
✔ working prompt testing environment

💡 ONE-LINE SUMMARY

Day 2 turns your Mac into a local multi-model AI reasoning system ready to be connected to your trading data pipeline.