Chains and agents

LLMOps Concepts

Max Knobbout, PhD

Applied Scientist, Uber

LLM lifecyle: Chains and agents

From prompts to applications

 

• To use this template, we need:
  • Examples
  • Input
• We go through a few steps:
  1. Receiving input
  2. Searching examples
  3. Prompt creation
  4. Output retrieval
  5. Output parsing

A chain using our template

 

The need for chains

 

 

 

• Develop sophisticated applications
• Establish a modular design, enhancing scalability and operational efficiency
• Unlock endless possibilities for customization

Agents

 

 

 

• Agents consist of:
  • Multiple actions (or tools)
  • An LLM deciding which action to take

 

• Useful when:
  • There are many actions
  • The optimal sequence of steps is unknown
  • We are uncertain about the inputs

Agents

Agents

Agents

The difference between chains and agents

 

 

	Chains 🔗	Agents 🤖
Nature	Deterministic	Adaptive
Complexity	Low	High
Flexibility	Low	High
Risk	Lower (due to predictability)	Higher (due to adaptability)

The development cycle

The development cycle

The development cycle

Let's practice!

LLMOps Concepts