What are components of Semantic kernel #9415
-
|
The components of the Semantic kernel are not mentioned in an official document of the semantic kernel. |
Beta Was this translation helpful? Give feedback.
Replies: 6 comments 3 replies
-
|
Hello @rakshahulle, have you taken a look at our Microsoft Learn site documentation? |
Beta Was this translation helpful? Give feedback.
-
|
Yes
…On Thu, 24 Oct 2024 at 7:08 PM, Evan Mattson ***@***.***> wrote:
Hello @rakshahulle <https://github.com/rakshahulle>, have you taken a
look at our Microsoft Learn site documentation
<https://learn.microsoft.com/en-us/semantic-kernel/overview/>?
—
Reply to this email directly, view it on GitHub
<#9415 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AXCZPRL6YUHLBZTX6GZI7JTZ5DZ65AVCNFSM6AAAAABQQE6T76VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMBUGEZTOMY>
.
You are receiving this because you were mentioned.Message ID:
***@***.***
com>
|
Beta Was this translation helpful? Give feedback.
-
|
Are concepts components? I have a dout
On Thu, 24 Oct 2024 at 7:15 PM, Raksha Hulle ***@***.***>
wrote:
… Yes
On Thu, 24 Oct 2024 at 7:08 PM, Evan Mattson ***@***.***>
wrote:
> Hello @rakshahulle <https://github.com/rakshahulle>, have you taken a
> look at our Microsoft Learn site documentation
> <https://learn.microsoft.com/en-us/semantic-kernel/overview/>?
>
> —
> Reply to this email directly, view it on GitHub
> <#9415 (comment)>,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/AXCZPRL6YUHLBZTX6GZI7JTZ5DZ65AVCNFSM6AAAAABQQE6T76VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMBUGEZTOMY>
> .
> You are receiving this because you were mentioned.Message ID:
> ***@***.***
> .com>
>
|
Beta Was this translation helpful? Give feedback.
-
|
Yes, the concepts show the components. For example, the central Kernel object where one specifies different AI services to interact with an LLM. One can bring their own native code as a Plugin or create a Plugin based on a prompt. Is there a specific question you have that we can help answer? |
Beta Was this translation helpful? Give feedback.
-
|
So, Kenel, AI service, Memory, Plugin, Planner , Prompt Engineering are
components of semantic kernel? Am I right? Also want to know about
Connectors, Semantic Function, Filters
…On Thu, 24 Oct 2024 at 7:52 PM, Evan Mattson ***@***.***> wrote:
Yes, the concepts show the components. For example, the central Kernel
<https://learn.microsoft.com/en-us/semantic-kernel/concepts/kernel?pivots=programming-language-python>
object where one specifies different AI services
<https://learn.microsoft.com/en-us/semantic-kernel/concepts/ai-services/>
to interact with an LLM. One can bring their own native code as a Plugin
<https://learn.microsoft.com/en-us/semantic-kernel/concepts/plugins/?pivots=programming-language-python>
or create a Plugin based on a prompt.
Is there a specific question you have that we can help answer?
—
Reply to this email directly, view it on GitHub
<#9415 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AXCZPRIGTH33VMID6TQXV2LZ5D7DNAVCNFSM6AAAAABQQE6T76VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMBUGE4TINQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***
com>
|
Beta Was this translation helpful? Give feedback.
-
|
I would say the highest level components are:
Then Plugins:
Planners:
Personas:
I'd say prompt engineer fits into the "plugin," "planner," or even the "persona" concept. It can take a bit to make sure one engineers the prompt effectively so they achieve their desired results. Same with filters -- filters provide an aspect to "hook" into certain places to the code -- like to view the final rendered prompt, or to get the function result from an auto function calling loop. One last important one is Telemetry: we use OpenTelemetry to allow developers to get an end-to-end view of their Semantic Kernel agent or operation. One can view the prompt, the number of tokens used, and more. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you so much for your answer; it is so useful. But, I am now a
little bit confused with the component Persona. As there is no physical
component of persona(Is Agent persona) or Can We say Persona as Prompt.
…On Thu, Oct 24, 2024 at 8:20 PM Evan Mattson ***@***.***> wrote:
I would say the highest level components are:
1. Plugins
2. Planners
3. Personas
Then Plugins:
- Plugin from a method (native code like C#, Python or Java)
- Plugin from a prompt (a natural language plugin to interact with an
LLM)
Planners:
- Interacting with the model to satisfy a user query/goal
- This can be done using one (or both) of the plugins from above.
- One may use "auto function calling" to have the model grounded,
as required by the developer
- When interacting with the model, one can specify the components:
Connectors (either memory or AI service like Azure OpenAI or any other
available AI connector)
Personas:
- You can construct various personas based on your use-case/scenario.
For example, I can create an "agent" to work a particular task. This could
be a "travel agent" or a "reviewing agent." (This doesn't necessarily mean
you use our Agent Framework, but that's possible, too.) As part of the
persona one can bring in the plugins and planners aspect. I'm not
specifically saying to use one of our Planners, but think about how to
"plan" a task for the LLM agent like "Find me my next open slot on my
calendar and book a lunch for me and my colleague." The LLM will break that
down into a plan like: "Find the next open slot on Evan's calendar (using a
plugin / auto function calling)" and then "using the result from the first
plugin, book a lunch for Evan and his colleague (using another plugin /
auto function calling)."
I'd say prompt engineer fits into the "plugin," "planner," or even the
"persona" concept. It can take a bit to make sure one engineers the prompt
effectively so they achieve their desired results. Same with filters --
filters provide an aspect to "hook" into certain places to the code -- like
to view the final rendered prompt, or to get the function result from an
auto function calling loop.
One last important one is Telemetry: we use OpenTelemetry to allow
developers to get an end-to-end view of their Semantic Kernel agent or
operation. One can view the prompt, the number of tokens used, and more.
—
Reply to this email directly, view it on GitHub
<#9415 (reply in thread)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AXCZPROH52CQ57K26N4JLODZ5ECM7AVCNFSM6AAAAABQQE6T76VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCMBUGIZDIOI>
.
You are receiving this because you were mentioned.Message ID:
***@***.***
com>
|
Beta Was this translation helpful? Give feedback.
-
|
You're welcome! I'm glad it helped. Regarding the "persona" component: it’s more of a guiding concept than a physical entity. It defines the style, behavior, or focus of the AI Agent during interactions. Think of it like a character or role that the agent adopts, shaping how it responds to users. While a prompt is a specific instruction, a persona is an ongoing set of preferences that influences how those prompts are interpreted. So, a persona isn't a prompt itself but rather a guiding framework that makes the agent’s interactions feel consistent and tailored to user needs. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks a lot.
…On Mon, 28 Oct 2024 at 8:12 PM, Sophia Lagerkrans-Pandey < ***@***.***> wrote:
Closed #9415
<#9415> as
resolved.
—
Reply to this email directly, view it on GitHub
<#9415>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AXCZPRKNYOTZ27M6JU67Y4TZ5ZENVAVCNFSM6AAAAABQQE6T76VHI2DSMVQWIX3LMV45UABFIRUXGY3VONZWS33OIV3GK3TUHI5E433UNFTGSY3BORUW63R3GE2TONRZGQ3A>
.
You are receiving this because you were mentioned.Message ID:
<microsoft/semantic-kernel/repo-discussions/9415/discussion_event/1576946@
github.com>
|
Beta Was this translation helpful? Give feedback.
Hello @rakshahulle, have you taken a look at our Microsoft Learn site documentation?