LLM's Multi-Model Orchestration

🤖 Multi-Model Orchestration

10. Ensemble LLMs
Route or combine responses from multiple models.
Use cases:
Use smaller LLM for easy questions, big one for complex
Voting / ranking model outputs
Safety filters (model A generates, model B critiques)


11. Soft Prompting, Prompt Tuning, Routing
Customize model behavior without retraining.
Soft Prompting: Add learned vectors before real prompt (e.g., prefix tuning)
Prompt Tuning: Trainable prompts on downstream task
Routing: Dynamic prompt switching depending on user intent