Model Inventory
Quick Reference
| Model Name | Type | Size | Purpose | Location |
|---|---|---|---|---|
| Image Models | Various | 149GB | Image generation, vision | models/image-models/ |
| Text Models | LLMs | 122GB | Text generation, NLP | models/text-models/ |
| Local Learning | 120B | TBD | 24/7 learning system | training/local-learning-system/ |
Image Models (149GB)
Source: ComfyUI installation
Location: models/image-models/ → /home/adminator/comfy/ComfyUI/image-models
Used by:
- Vision agents
- Image generation agents
- Image analysis agents
Note: Symlinked to avoid duplication. Do NOT delete source!
Text Models (122GB)
Source: Llama.cpp installation
Location: models/text-models/ → /home/adminator/llama.cpp/text-models
Used by:
- NLP agents
- Text generation agents
- Sentiment analyzers
- Intent classifiers
Note: Symlinked to avoid duplication. Do NOT delete source!
Local Learning System
Location: training/local-learning-system/
From expert_agents/LOCAL_MODEL_AGENT.md:
- 24/7 learning capability
- Knowledge accumulation
- Offline inference
- Custom domain learning
Integration: Connect this to SMO for continuous learning
Specialized Systems
Vision System
Computer vision components and pipelines.
Arthur
Specialized ML system (investigate purpose).
Model Armour
Security and validation tools for models.
Creating Agent Models
Micro-agents need small, fast models:
- Start with large model (from text-models/)
- Fine-tune on specific task
- Quantize (FP32 → INT8)
- Optimize for latency
- Store in
models/agent-models/
Target specs:
- Size: <500MB
- Latency: <100ms
- Accuracy: >90%
Next Steps
- Catalog all models in image-models/
- Catalog all models in text-models/
- Create agent-specific models
- Set up MLflow registry
- Build training pipeline