[Python] Conversation summary skill porting part I (microsoft#853)

### Motivation and Context
Porting conversation summary skill. Integration test is added as well.
Action item and topic will be ported later.

---------

Co-authored-by: Po-Wei Huang <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Devis Lucato <[email protected]>
Co-authored-by: Devis Lucato <[email protected]>

FEATURE_MATRIX.md

@@ -28,7 +28,7 @@
 | | C# | Python | Notes |
 |---|---|---|---|
 | TextMemorySkill                   | ✅ | ✅ | |
-| ConversationSummarySkill          | ✅ | ❌ | |
+| ConversationSummarySkill          | ✅ | ✅ | |
 | FileIOSkill                       | ✅ | ✅ | |
 | HttpSkill                         | ✅ | ✅ | |
 | MathSkill                         | ✅ | ✅ | |

python/semantic_kernel/core_skills/__init__.py

@@ -1,5 +1,8 @@
 # Copyright (c) Microsoft. All rights reserved.
 
+from semantic_kernel.core_skills.conversation_summary_skill import (
+    ConversationSummarySkill,
+)
 from semantic_kernel.core_skills.file_io_skill import FileIOSkill
 from semantic_kernel.core_skills.http_skill import HttpSkill
 from semantic_kernel.core_skills.math_skill import MathSkill
@@ -14,5 +17,6 @@
     "TimeSkill",
     "HttpSkill",
     "BasicPlanner",
+    "ConversationSummarySkill",
     "MathSkill",
 ]

python/semantic_kernel/core_skills/conversation_summary_skill.py

@@ -0,0 +1,63 @@
+# Copyright (c) Microsoft. All rights reserved.
+
+from semantic_kernel.kernel import Kernel
+from semantic_kernel.orchestration.sk_context import SKContext
+from semantic_kernel.skill_definition import sk_function
+from semantic_kernel.text import text_chunker
+from semantic_kernel.text.function_extension import aggregate_chunked_results_async
+
+
+class ConversationSummarySkill:
+    """
+    Semantic skill that enables conversations summarization.
+    """
+
+    # The max tokens to process in a single semantic function call.
+    _max_tokens = 1024
+
+    _summarize_conversation_prompt_template = (
+        "BEGIN CONTENT TO SUMMARIZE:\n"
+        "{{" + "$INPUT" + "}}\n"
+        "END CONTENT TO SUMMARIZE.\n"
+        "Summarize the conversation in 'CONTENT TO SUMMARIZE',\
+            identifying main points of discussion and any conclusions that were reached.\n"
+        "Do not incorporate other general knowledge.\n"
+        "Summary is in plain text, in complete sentences, with no markup or tags.\n"
+        "\nBEGIN SUMMARY:\n"
+    )
+
+    def __init__(self, kernel: Kernel):
+        self._summarizeConversationFunction = kernel.create_semantic_function(
+            ConversationSummarySkill._summarize_conversation_prompt_template,
+            skill_name=ConversationSummarySkill.__name__,
+            description="Given a section of a conversation transcript, summarize the part of the conversation.",
+            max_tokens=ConversationSummarySkill._max_tokens,
+            temperature=0.1,
+            top_p=0.5,
+        )
+
+    @sk_function(
+        description="Given a long conversation transcript, summarize the conversation.",
+        name="SummarizeConversation",
+        input_description="A long conversation transcript.",
+    )
+    async def summarize_conversation_async(
+        self, input: str, context: SKContext
+    ) -> SKContext:
+        """
+        Given a long conversation transcript, summarize the conversation.
+
+        :param input: A long conversation transcript.
+        :param context: The SKContext for function execution.
+        :return: SKContext with the summarized conversation result.
+        """
+        lines = text_chunker._split_text_lines(
+            input, ConversationSummarySkill._max_tokens, True
+        )
+        paragraphs = text_chunker._split_text_paragraph(
+            lines, ConversationSummarySkill._max_tokens
+        )
+
+        return await aggregate_chunked_results_async(
+            self._summarizeConversationFunction, paragraphs, context
+        )

python/tests/integration/completions/e2e_text_completion.py

@@ -1,7 +1,12 @@
+# Copyright (c) Microsoft. All rights reserved.
+
 import logging
 import time
 
 import semantic_kernel as sk
+from semantic_kernel.core_skills.conversation_summary_skill import (
+    ConversationSummarySkill,
+)
 
 logging.basicConfig(level=logging.DEBUG)
 logger = logging.getLogger()
@@ -316,3 +321,55 @@ async def simple_completion(kernel: sk.Kernel):
         f"Completion using context, additional variables, and additional string: '{output}'"
     )
     assert len(output) > 0
+
+
+async def summarize_conversation_using_skill(kernel: sk.Kernel):
+    ChatTranscript = """John: Hello, how are you?
+        Jane: I'm fine, thanks. How are you?
+        John: I'm doing well, writing some example code.
+        Jane: That's great! I'm writing some example code too.
+        John: What are you writing?
+        Jane: I'm writing a chatbot.
+        John: That's cool. I'm writing a chatbot too.
+        Jane: What language are you writing it in?
+        John: I'm writing it in C#.
+        Jane: I'm writing it in Python.
+        John: That's cool. I need to learn Python.
+        Jane: I need to learn C#.
+        John: Can I try out your chatbot?
+        Jane: Sure, here's the link.
+        John: Thanks!
+        Jane: You're welcome.
+        Jane: Look at this poem my chatbot wrote:
+        Jane: Roses are red
+        Jane: Violets are blue
+        Jane: I'm writing a chatbot
+        Jane: What about you?
+        John: That's cool. Let me see if mine will write a poem, too.
+        John: Here's a poem my chatbot wrote:
+        John: The signularity of the universe is a mystery.
+        Jane: You might want to try using a different model.
+        John: I'm using the GPT-2 model. That makes sense.
+        John: Here is a new poem after updating the model.
+        John: The universe is a mystery.
+        John: The universe is a mystery.
+        John: The universe is a mystery.
+        Jane: Sure, what's the problem?
+        John: Thanks for the help!
+        Jane: I'm now writing a bot to summarize conversations.
+        Jane: I have some bad news, we're only half way there.
+        John: Maybe there is a large piece of text we can use to generate a long conversation.
+        Jane: That's a good idea. Let me see if I can find one. Maybe Lorem Ipsum?
+        John: Yeah, that's a good idea."""
+
+    conversationSummarySkill = kernel.import_skill(
+        ConversationSummarySkill(kernel), "conversationSummary"
+    )
+    summary = await kernel.run_async(
+        conversationSummarySkill["SummarizeConversation"], input_str=ChatTranscript
+    )
+
+    output = str(summary).strip().lower()
+    print(output)
+    assert "john" in output and "jane" in output
+    assert len(output) < len(ChatTranscript)

python/tests/integration/completions/test_conversation_summary_skill.py

@@ -0,0 +1,55 @@
+# Copyright (c) Microsoft. All rights reserved.
+
+import asyncio
+import os
+
+import e2e_text_completion
+import pytest
+
+import semantic_kernel as sk
+import semantic_kernel.connectors.ai.open_ai as sk_oai
+
+
+@pytest.mark.asyncio
+async def test_azure_summarize_conversation_using_skill():
+    kernel = sk.Kernel()
+
+    if "Python_Integration_Tests" in os.environ:
+        deployment_name = os.environ["AzureOpenAI__DeploymentName"]
+        api_key = os.environ["AzureOpenAI__ApiKey"]
+        endpoint = os.environ["AzureOpenAI__Endpoint"]
+    else:
+        # Load credentials from .env file
+        deployment_name, api_key, endpoint = sk.azure_openai_settings_from_dot_env()
+        deployment_name = "text-davinci-003"
+
+    kernel.add_text_completion_service(
+        "text_completion",
+        sk_oai.AzureTextCompletion(deployment_name, endpoint, api_key),
+    )
+
+    await e2e_text_completion.summarize_conversation_using_skill(kernel)
+
+
+@pytest.mark.asyncio
+async def test_oai_summarize_conversation_using_skill():
+    kernel = sk.Kernel()
+
+    if "Python_Integration_Tests" in os.environ:
+        api_key = os.environ["OpenAI__ApiKey"]
+        org_id = None
+    else:
+        # Load credentials from .env file
+        api_key, org_id = sk.openai_settings_from_dot_env()
+
+    kernel.add_chat_service(
+        "davinci-003",
+        sk_oai.OpenAITextCompletion("text-davinci-003", api_key, org_id=org_id),
+    )
+
+    await e2e_text_completion.summarize_conversation_using_skill(kernel)
+
+
+if __name__ == "__main__":
+    asyncio.run(test_oai_summarize_conversation_using_skill())
+    asyncio.run(test_azure_summarize_conversation_using_skill())