Initial commit: 2125_NBB project
This commit is contained in:
Executable
+194
@@ -0,0 +1,194 @@
|
||||
## 5. Usage Guide (Summary)
|
||||
|
||||
See `DUAL_LAYER_USAGE_GUIDE.md` in Part 2 for the complete guide. Key commands:
|
||||
|
||||
### Start Server
|
||||
```bash
|
||||
python3 /home/dave/server.py
|
||||
```
|
||||
|
||||
### Access Dashboard
|
||||
Open in browser: **http://localhost:8000/glyphs/index.html**
|
||||
|
||||
### Test Symbolic Endpoints
|
||||
```bash
|
||||
# Check status
|
||||
curl http://localhost:8000/api/symbolic/status
|
||||
|
||||
# Activate glyph
|
||||
curl -X POST http://localhost:8000/api/symbolic/activate \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{"intent": "I need primordial authority", "request_type": "chat"}'
|
||||
```
|
||||
|
||||
### Chat with Glyph Activation
|
||||
```bash
|
||||
curl -X POST http://localhost:8000/api/chat \
|
||||
-H "Content-Type: application/json" \
|
||||
-d '{
|
||||
"model": "llama-3.5-35b",
|
||||
"messages": [{"role": "user", "content": "Hello"}],
|
||||
"temperature": 0.7,
|
||||
"glyph_activation": {
|
||||
"intent": "I need root authority",
|
||||
"request_type": "chat"
|
||||
}
|
||||
}'
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 6. Testing & Validation
|
||||
|
||||
### Dual-Layer Endpoint Tests
|
||||
|
||||
All 5 symbolic API endpoints verified via TestClient returning 200 OK:
|
||||
|
||||
| Endpoint | Method | Status |
|
||||
|----------|--------|--------|
|
||||
| /api/symbolic/status | GET | Verified |
|
||||
| /api/symbolic/glyphs | GET | Verified |
|
||||
| /api/symbolic/activate | POST | Verified |
|
||||
| /api/symbolic/deactivate | POST | Verified |
|
||||
| /api/symbolic/routing/summary | GET | Verified |
|
||||
|
||||
### Multi-Glyph Resonance Tests
|
||||
|
||||
The validation suite (test_multi_glyph_resonance.py) runs 12 tests:
|
||||
1. New operations in OP_TABLE
|
||||
2. XICContext.glyph_contexts field
|
||||
3. PUSH_GLYPH_CONTEXT accumulation
|
||||
4. CLEAR_GLYPH_CONTEXT reset
|
||||
5. Guardrail enforcement
|
||||
6. run_symbolic_pipeline signature
|
||||
7. compute_multi_glyph_resonance() exists
|
||||
8. Multi-glyph computation structure
|
||||
9. execute_symbolic with glyph_ids
|
||||
10. Backward compatibility
|
||||
11. Demo programs validation
|
||||
12. Multi-glyph demo structure
|
||||
|
||||
Run with:
|
||||
```bash
|
||||
python3 /home/dave/superdave/test_multi_glyph_resonance.py
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
## 7. Key Decisions Log
|
||||
|
||||
### Architecture
|
||||
|
||||
| Decision | Rationale |
|
||||
|----------|-----------|
|
||||
| Dual-layer design | Separates symbolic intent (glyphs/resonance) from computational execution (models/VRAM) |
|
||||
| Async VRAM lock (asyncio.Lock) | Prevents threading lock timeouts while keeping concurrent safety |
|
||||
| Priority formula: min(10.0, power_boost / 40.0) | G001 (387.95x) gets max priority 10.0; normal glyphs scale proportionally |
|
||||
| Resonance formula: 40% activation + 30% frequency + 30% symbolic | Balances power count, boost intensity, and type significance on 0-100 scale |
|
||||
| Singleton managers (get_vram_manager, get_symbolic_engine) | Ensures global state consistency across requests |
|
||||
|
||||
### VRAM Management
|
||||
|
||||
| Decision | Rationale |
|
||||
|----------|-----------|
|
||||
| 8GB GTX1080 limits | Hardware constraint; Warning=6.5GB, Critical=7.5GB |
|
||||
| Forge/Janus mutex | Both consume ~4.5-5GB each; running together would exceed 8GB |
|
||||
| Priority-based deactivation | Higher-priority glyphs can evict lower-priority ones when VRAM is full |
|
||||
| Per-type VRAM budgets | Different model types need different amounts (llama=2GB, forge=4.5GB, janus=5GB) |
|
||||
|
||||
### Glyph Configuration
|
||||
|
||||
| Decision | Rationale |
|
||||
|----------|-----------|
|
||||
| G001 gets 152 superpowers, priority 10.0 | Root glyph (Ledo) has maximum authority |
|
||||
| G002-G600 get 5-25 powers dynamically | Metric-based assignment ensures balanced distribution |
|
||||
| 9 specialized types | Maps to available computational models (Llama, Forge, etc.) |
|
||||
| aether_node: 7.5GB max VRAM | G001 needs maximum resources for full capability |
|
||||
|
||||
### Specialized Type Routing
|
||||
|
||||
| Decision | Rationale |
|
||||
|----------|-----------|
|
||||
| star_bloom_creativity -> forge | Creative tasks map to image generation |
|
||||
| frost_steel_stabilizer -> llama | Safety/stability handled by LLM constraints |
|
||||
| mirror_weave_reasoning -> llama | Reasoning/logic handled by LLM enhancements |
|
||||
| monument_grade_equilibrium -> llama | System balance handled by LLM orchestration |
|
||||
|
||||
### Dashboard Design
|
||||
|
||||
| Decision | Rationale |
|
||||
|----------|-----------|
|
||||
| Real-time VRAM bar | Visual indicator of GPU memory pressure |
|
||||
| 5-second auto-refresh | Balance between responsiveness and API load |
|
||||
| Activity log with 20 entries | Keep UI clean while showing recent history |
|
||||
| Fixed refresh button | Always accessible for manual refresh |
|
||||
|
||||
---
|
||||
|
||||
## 8. Next Steps
|
||||
|
||||
### Immediate (Blocked)
|
||||
1. **Fix server persistence**: Server keeps shutting down when shell times out
|
||||
- Use: `nohup python3 /home/dave/server.py &`
|
||||
- Or set up systemd/tmux service
|
||||
2. **Verify dashboard** from browser at http://localhost:8000/glyphs/index.html
|
||||
|
||||
### Short Term
|
||||
3. Test glyph activation from dashboard UI
|
||||
4. Connect to Pinokio for real model execution
|
||||
5. Verify VRAM monitoring during operation
|
||||
|
||||
### Medium Term
|
||||
6. Connect Llama Chat (Tabby API endpoint)
|
||||
7. Connect Forge Image Generation (diffusers/SDXL-Turbo)
|
||||
8. Connect Google AI Vision (Gemini API key)
|
||||
9. Connect Janus Video Generation
|
||||
|
||||
### Long Term
|
||||
10. Convert to single EXE via PyInstaller
|
||||
11. Tune routing thresholds based on real usage
|
||||
12. Expand beyond 600 glyphs if desired
|
||||
|
||||
---
|
||||
|
||||
## 9. Session Context & Thinking Log
|
||||
|
||||
### Build Sequence
|
||||
1. Created dual-layer package structure (dual_layer/__init__.py, router.py, vram_manager.py, symbolic_engine.py)
|
||||
2. Built routing system mapping 9 specialized types to models with constraints/enhancements
|
||||
3. Implemented async VRAM manager with Forge/Janus mutex and priority deactivation
|
||||
4. Created SymbolicEngine for intent-based glyph activation and resonance calculation
|
||||
5. Built FastAPI integration with 5 symbolic endpoints
|
||||
6. Created real-time glyph activation dashboard (HTML/CSS/JS)
|
||||
7. Implemented glyph-enhanced model execution (constraint/enhancement framework)
|
||||
8. Enhanced server.py with dual-layer import and dashboard mounting
|
||||
9. Fixed bugs: threading.Lock -> asyncio.Lock, import paths, parameter mismatches
|
||||
10. Verified all 5 endpoints via TestClient returning 200 OK
|
||||
11. Created comprehensive usage guide
|
||||
|
||||
### Bugs Fixed
|
||||
- **Threading lock timeout**: Replaced threading.Lock() with asyncio.Lock() in VRAMManager
|
||||
- **Import paths**: Standardized "dual_layer.*" -> "superdave.dual_layer.*"
|
||||
- **GlyphActivationEvent parameter mismatch**: Changed success/failure_reason to context dict
|
||||
- **Variable naming conflict**: Fixed logger redefinition in server.py (line 37)
|
||||
|
||||
### Critical Rules Enforced
|
||||
- NEVER run Forge + Janus simultaneously (8GB VRAM crash risk)
|
||||
- G001 has maximum priority (10.0) and VRAM budget (7.5GB)
|
||||
- Priority-based preemption for VRAM allocation
|
||||
- FedMart telemetry on every glyph activation
|
||||
|
||||
---
|
||||
|
||||
## 10. Version History
|
||||
|
||||
| Date | Version | Changes |
|
||||
|------|---------|---------|
|
||||
| Sat Jun 13 2026 | 1.0.0 | Initial dual-layer build complete |
|
||||
| Sat Jun 13 2026 | 1.0.1 | Fixed async lock, imports, event params |
|
||||
| Sat Jun 13 2026 | 1.0.2 | Added dashboard, enhanced server.py |
|
||||
| Sat Jun 13 2026 | 1.0.3 | Endpoint verification, usage guide |
|
||||
|
||||
---
|
||||
|
||||
*End of Session Export*
|
||||
Reference in New Issue
Block a user