🍓 Strawberry Model System Flowchart 🍓
Input Query
Large Language Model (LLM)
STaR: Self-Taught Reasoner
Dynamic Self-Discovery
RAG Optimization
DSPy Integration
Graph of Thoughts (GoT)
Monte Carlo Tree Search (MCTS)
DSPy/CLIN Tooling
Self-Play Optimization
Graph-based Knowledge Base
GraphRAG-inspired Retrieval
Dynamic Reasoning Modules
Continuous Learning Loop
Updated Knowledge Base
Refined Model
Output
New Data Collection
Pretraining with New Data
×
Input:
Output:
🍓 Explanation 🍓
- Input and Initial Processing
- Advanced Reasoning and Optimization
- Knowledge Management and Retrieval
- Continuous Learning and Refinement
- Data Collection and Model Updating
Note:
🍓 Red box: The core LLM component
🍓 Pink boxes: Steps where the LLM's weights are further trained or its knowledge database is updated
🍓 White box: New data collection for pretraining