Mythos Local - Simple Start Guide

Status: ✅ FIXED & VERIFIED

The configuration has been completely rewritten with correct model URLs.

🚀 Quickest Start (3 commands)

cd ~/mythos_local/amd-vulkan-llm-project
source venv/bin/activate
sudo bash run_chat.sh

That's it! The model will auto-download on first run.

📋 What's Available

Option 1: Terminal Chat (Recommended First)

sudo bash run_chat.sh

• Clean terminal interface
• Streaming responses
• Commands like /help, /benchmark, /temp 0.9

Option 2: Web Interface (Pretty UI)

sudo bash run_web.sh

• Opens at http://localhost:7860
• Sliders for temperature, top-p, etc.
• Upload documents for RAG

Option 3: Just Download Model

source venv/bin/activate
python3 test_download.py

Option 4: Verify Everything First

source venv/bin/activate
./verify_setup.sh

🔍 Verify Config is Correct

./show_config.sh

You should see:

✓✓✓ CORRECT - Using bartowski repos ✓✓✓

NOT:

❌❌❌ WRONG - NOT using bartowski! ❌❌❌

⚡ Performance Expectations

First Run: - Download: 5-10 minutes (~4.7GB) - Model loading: 15-20 seconds - Then ready to chat!

Subsequent Runs: - Model loading: 10-15 seconds - Ready immediately

Speed: - CPU-only: 5-15 tokens/second - With Vulkan: 15-30 tokens/second - RAM usage: 6-10GB

🎯 What Was Fixed

1. config.yaml - Completely rewritten
2. ✅ Now uses: bartowski/Qwen2.5-7B-Instruct-GGUF

3. ❌ Was using: Qwen/Qwen2.5-7B-Instruct-GGUF (404 error)

4. Added verification scripts:

5. show_download_url.py - Shows exact download URL
6. debug_config.py - Debug config loading
7. test_download.py - Test download only
8. verify_setup.sh - Full verification

9. show_config.sh - Quick config check

10. Added simple runners:

11. run_chat.sh - One-command start for terminal
12. run_web.sh - One-command start for web UI

🐛 Troubleshooting

Still seeing 404 errors?

1. Check you're in the right directory: bash pwd # Should show: /home/a4/mythos_local/amd-vulkan-llm-project

2. Verify config: bash grep "repo_id" config.yaml | head -1 # Should show: repo_id: "bartowski/Qwen2.5-7B-Instruct-GGUF"

3. Clear Python cache: bash rm -rf __pycache__ engine/__pycache__ ui/__pycache__

4. Show actual download URL: bash python3 show_download_url.py # Should show "bartowski" and "✓✓✓ CORRECT"

📚 Documentation Files

• FINAL_INSTRUCTIONS.txt - Most detailed guide
• START_HERE.txt - Comprehensive start guide
• FIXED_README.txt - What was fixed and why
• README.md - Full technical documentation
• README_SIMPLE.md - This file (simple guide)

🎓 Chat Commands (Terminal Mode)

Once running, type these:

• /help - Show all commands
• /benchmark - Test model quality
• /system creative - Creative writing mode
• /system coding - Coding assistant mode
• /temp 0.9 - More creative responses
• /reflect on - Better quality (slower)
• /quit - Exit

✅ Success Checklist

Before running, verify:

• [ ] You're in ~/mythos_local/amd-vulkan-llm-project
• [ ] Virtual environment is activated (venv)
• [ ] ./show_config.sh shows "bartowski"
• [ ] You have ~6GB free disk space for model
• [ ] You have ~10GB free RAM

Then run: ./run_chat.sh

🚀 Ready!

cd ~/mythos_local/amd-vulkan-llm-project
source venv/bin/activate
./run_chat.sh

The first run will download the model. Be patient!

Enjoy your Mythos-tier local LLM! 🎉