Core Plugins

in Semantic Kernel

Jan Kirenz

Core Plugins

  • We want to have a vat of plugins … and then find the right plugin to fit the goal …

  • You can find more about the predefined plugins used below here.

Setup

Python

import semantic_kernel as sk
from semantic_kernel.connectors.ai.open_ai import OpenAIChatCompletion, OpenAITextEmbedding
from semantic_kernel.connectors.ai.open_ai import AzureChatCompletion, AzureTextEmbedding
from semantic_kernel.planning import ActionPlanner
from semantic_kernel.planning import SequentialPlanner
from semantic_kernel.core_skills.text_skill import TextSkill
from semantic_kernel.planning import SequentialPlanner
from semantic_kernel.core_skills.text_skill import TextSkill
from semantic_kernel.core_skills import FileIOSkill, MathSkill, TextSkill, TimeSkill

from IPython.display import display, Markdown

Kernel

kernel = sk.Kernel()

api_key, org_id = sk.openai_settings_from_dot_env()

kernel.add_text_completion_service(
    "openaicompletion", OpenAIChatCompletion("gpt-3.5-turbo", api_key, org_id))

kernel.add_text_embedding_generation_service(
    "openaiembedding", OpenAITextEmbedding("text-embedding-ada-002", api_key, org_id))

Add Core Plugins

Import plugins

  • Action Planner is a simple tool which can find the right tool to get things done
planner = ActionPlanner(kernel)
  • Adding the tools for the kernel to do math, to read/write files, to tell the time, and to play with text:
kernel.import_skill(MathSkill(), "math")
kernel.import_skill(FileIOSkill(), "fileIO")
kernel.import_skill(TimeSkill(), "time")
kernel.import_skill(TextSkill(), "text")

Example Plugins

Math

ask = "What is the sum of 110 and 990?"
  • Finding the most similar function available to get that done…
plan = await planner.create_plan_async(goal=ask)
print(
    f" The best single function to use is `{plan._skill_name}.{plan._function.name}`")

Date

ask = "What is today?"
print(f" Finding the most similar function available to get that done...")
plan = await planner.create_plan_async(goal=ask)
print(
    f" The best single function to use is `{plan._skill_name}.{plan._function.name}`")

Text

ask = "How do I write the word 'text' to a file?"
print(f" Finding the most similar function available to get that done...")
plan = await planner.create_plan_async(goal=ask)
print(
    f" The best single function to use is `{plan._skill_name}.{plan._function.name}`")

Text example Poem

Custom Plugin LiterateFriend

plugins_directory = "../plugins-sk"
writer_plugin = kernel.import_semantic_skill_from_directory(
    plugins_directory, "LiterateFriend")

Use Sequential Planner

planner = SequentialPlanner(kernel)

Task

ask = """
Tomorrow is Valentine's day. I need to come up with a poem. Translate the poem to German.
"""
  • This will require two plugins (write poem and translate)

Call planner

plan = await planner.create_plan_async(goal=ask)

Obtain result

result = await plan.invoke_async()

Output

for index, step in enumerate(plan._steps):
    print(f"✅ Step {index+1} used function `{step._function.name}`")

trace_resultp = True

display(
    Markdown(f"## ✨ Generated result from the ask: {ask}\n\n---\n" + str(result)))

Display output

✅ Step 1 used function `ShortPoem`
✅ Step 2 used function `Translate`
✨ Generated result from the ask:

Tomorrow is Valentine's day. I need to come up with a poem. Translate the poem to German.

Rosen sind rot, Veilchen sind blau, Valentinstag ist da, was wirst du tun? Kaufe Schokolade und Blumen oder schreibe einen Liebesbrief, Vergiss nur nicht, es alles besser zu machen.

Wenn du Single bist, keine Sorge, Gönn dir selbst, zu...
  • mmmhhh, maybe lost in translation 🙂

Take a closer look at the output

for index, step in enumerate(plan._steps):
    print(f"✅ Step {index+1} used function `{step._function.name}`")

trace_resultp = True

if trace_resultp:
    print("Longform trace:\n")
    for index, step in enumerate(plan._steps):
        print("Step:", index)
        print("Description:", step.description)
        print("Function:", step.skill_name + "." + step._function.name)
        print("Output vars:", step._outputs)
        if len(step._outputs) > 0:
            print("  Output:\n", str.replace(
                result[step._outputs[0]], "\n", "\n  "))

display(
    Markdown(f"## ✨ Generated result from the ask: {ask}\n\n---\n" + str(result)))

Take a closer look at the output


✅ Step 1 used function `ShortPoem`
✅ Step 2 used function `Translate`
Longform trace:

Step: 0
Description: Turn a scenario into a short and entertaining poem.
Function: LiterateFriend.ShortPoem
Output vars: ['POEM']
  Output:
 Roses are red, violets are blue,
  Valentine's Day is here, what will you do?
  Buy chocolates and flowers, or write a love letter,
  Just don't forget, to make it all better.
  
  If you're single, don't you fret,
  Treat yourself, to

Step: 1
Description: Translate the input into a language of your choice
Function: LiterateFriend.Translate
Output vars: ['RESULT__TRANSLATED_POEM']
  Output:
 Rosen sind rot, Veilchen sind blau,
  Valentinstag ist da, was wirst du tun?
  Kaufe Schokolade und Blumen oder schreibe einen Liebesbrief,
  Vergiss nur nicht, es alles besser zu machen.
  
  Wenn du Single bist, keine Sorge,
  Gönn dir selbst, zu...

Acknowledgments

This tutorial is mainly based on the excellent course “How Business Thinkers Can Start Building AI Plugins With Semantic Kernel” provided by John Maeda and Andrew Ng

What’s next?

Congratulations! You have completed this tutorial 👍

Next, you may want to go back to the lab’s website

Jan Kirenz