Skip to content

components.md

Latest commit

 

History

History
80 lines (57 loc) · 3.43 KB

components.md

File metadata and controls

80 lines (57 loc) · 3.43 KB

Component

Kernel

  • Kernel: "The core, center, or essence of an object or system." —Wiktionary
  • Kernel refers to an instance of the processing engine that processes an ASK all the way through to fulfillment.
using Microsoft.SemanticKernel;
var myKernel = Kernel.Builder.Build();

There are a variety of things that you can do with myKernel that include:

  • Configuring the kernel to use OpenAI or Azure OpenAI
  • Sourcing a collection of skills
  • Chaining multiple skills' together
  • Customize how the kernel works to fit your exact needs

Planner

  • The planner works backwards from a goal that’s provided from a user's ASK.
  • The planner has access to either a pre-defined library of pre-made skills and/or a dynamically defined set of skills it is able to fulfill an ASK with confidence.
  • The planner calls upon memories to best situate the ASK's context and connectors to call APIs and to leverage other external capabilities.
  • The planner will operate within the skills it has available. In the event that a desired skill does not exist, the planner can suggest you to create the skill. Or, depending upon the level of complexity the kernel can help you write the missing skill.

Skills

  • A skill refers to a domain of expertise made available to the kernel as a single function, or as a group of functions related to the skill.

A function is the basic building block for a skill. A function can be expressed as either:

  1. an LLM AI prompt — also called a "semantic" function
  2. native computer code -- also called a "native" function
  • A skill is the container in which functions live.
  • Skill is a directory with subdirectories of semantic and native functions
  • Native functions are loosely inspired by Azure Functions and exist as individual native skill files.
MyAppSource
│
└───MySkillsDirectory
    │
    └─── MySemanticSkill (a directory)
    |   │
    |   └─── MyFirstSemanticFunction (a directory)
    |   └─── MyOtherSemanticFunction (a directory)
    │
    └─── MyNativeSkill.cs (a file)
    └─── MyOtherNativeSkill.cs (a file)

Memories

Memories are like RAM where AI models are like CPU

3 ways SK accesses memories

  • KV Pairs: Similar to env variable, convenient lookup
  • Local storage: From local file
  • Semantic memory search: Access from embeddings

How does Semantic Memory Search work?

  • Embeddings are a way of representing words or other data as vectors in a high-dimensional space.
  • Similar words or data will have similar vectors, and different words or data will have different vectors.
  • So basically you take a sentence, paragraph, or entire page of text, and then generate the corresponding embedding vector.
  • When a query is performed, the query is transformed to its embedding representation, and then a search is performed through all the existing embedding vectors to find the most similar ones.

Why are embeddings important?

  • We need to consider the length of the input text based on the token limit of the model we choose to use
  • GPT-4 can handle up to 8,192 tokens per input, while GPT-3 can only handle up to 4,096 tokens.
  • Embeddings break down larger text into smaller pieces.
  • An embedding vector is like a compressed representation of the text that preserves its meaning and context.
  • Then we can compare the embedding vectors of our summaries with the embedding vector of our prompt and select the most similar ones.

Connectors