[PY] feat: add prompt manager (#1233)

## Linked issues

closes: #1065

## Details

1. Add `PromptManager` and corresponding classes
2. Accessing `DataSource.name` will throw error with `TextDataSource`
instance. Update `TextDataSource` and `DataSource` class to expose the
`name` property to fix this issue.
3. Add `from_dict` method to some dataclasses to allow them initialize
from a json file

Note: augmentations and actions related logic are not included in this
PR as they're not ready yet. I have added comments to related issues as
follow up items.

## Attestation Checklist

- [x] My code follows the style guidelines of this project

- I have checked for/fixed spelling, linting, and other errors
- I have commented my code for clarity
- I have made corresponding changes to the documentation (updating the
doc strings in the code is sufficient)
- My changes generate no new warnings
- I have added tests that validates my changes, and provides sufficient
test coverage. I have tested with:
  - Local testing
  - E2E testing in Teams
- New and existing unit tests pass locally with my changes

---------

Co-authored-by: Alex Acebo <[email protected]>

Author: blackchoey

python/packages/ai/teams/ai/data_sources/data_source.py

@@ -17,8 +17,10 @@ class DataSource(ABC):
     A data source that can be used to render text that's added to a prompt.
     """
 
-    name: str
-    "Name of the data source."
+    @property
+    @abstractmethod
+    def name(self) -> str:
+        "Name of the data source."
 
     @abstractmethod
     async def render_data(

python/packages/ai/teams/ai/data_sources/text_data_source.py

@@ -38,7 +38,7 @@ def __init__(self, name: str, text: str) -> None:
         self._text = text
 
     @property
-    def get_name(self) -> str:
+    def name(self) -> str:
         """
         Name of the data source.
         """

python/packages/ai/teams/ai/promptsv2/__init__.py

@@ -12,8 +12,13 @@
 from .group_section import GroupSection
 from .layout_engine import LayoutEngine
 from .message import ImageContentPart, ImageUrl, Message, TextContentPart
-from .prompt_section import PromptFunctions, PromptSection
+from .prompt import Prompt
+from .prompt_functions import PromptFunction, PromptFunctions
+from .prompt_manager import PromptManager
+from .prompt_manager_options import PromptManagerOptions
+from .prompt_section import PromptSection
 from .prompt_section_base import PromptSectionBase
+from .prompt_template import PromptTemplate
 from .rendered_prompt_section import RenderedPromptSection
 from .system_message import SystemMessage
 from .template_section import TemplateSection

python/packages/ai/teams/ai/promptsv2/completion_config.py

@@ -69,3 +69,20 @@ class CompletionConfig:
     stop_sequences: Optional[List[str]] = None
     temperature: float = 0
     top_p: float = 0
+
+    @classmethod
+    def from_dict(cls, data: dict) -> "CompletionConfig":
+        return cls(
+            completion_type=data.get("completion_type"),
+            frequency_penalty=data.get("frequency_penalty", 0),
+            include_history=data.get("include_history", True),
+            include_input=data.get("include_input", True),
+            include_images=data.get("include_images", False),
+            max_tokens=data.get("max_tokens", 150),
+            max_input_tokens=data.get("max_input_tokens", 2048),
+            model=data.get("model"),
+            presence_penalty=data.get("presence_penalty", 0),
+            stop_sequences=data.get("stop_sequences"),
+            temperature=data.get("temperature", 0),
+            top_p=data.get("top_p", 0),
+        )

python/packages/ai/teams/ai/promptsv2/prompt_manager.py

@@ -0,0 +1,301 @@
+"""
+Copyright (c) Microsoft Corporation. All rights reserved.
+Licensed under the MIT License.
+"""
+
+import json
+import os
+from copy import deepcopy
+from pathlib import Path
+from typing import Any, Dict, List
+
+from botbuilder.core import TurnContext
+
+from ...app_error import ApplicationError
+from ...state import Memory
+from ..data_sources import DataSource
+from ..tokenizers import Tokenizer
+from .conversation_history import ConversationHistory
+from .group_section import GroupSection
+from .prompt import Prompt
+from .prompt_functions import PromptFunction, PromptFunctions
+from .prompt_manager_options import PromptManagerOptions
+from .prompt_section import PromptSection
+from .prompt_template import PromptTemplate
+from .prompt_template_config import PromptTemplateConfig
+from .template_section import TemplateSection
+from .user_input_message import UserInputMessage
+from .user_message import UserMessage
+
+
+class PromptManager(PromptFunctions):
+    _options: PromptManagerOptions
+    _data_sources: Dict[str, DataSource]
+    _functions: Dict[str, PromptFunction]
+    _prompts: Dict[str, PromptTemplate]
+
+    def __init__(self, options: PromptManagerOptions):
+        """
+        Creates a new 'PromptManager' instance.
+
+        Args:
+            options (PromptManagerOptions): Options used to configure the prompt manager.
+        """
+        self._options = options
+        self._data_sources = {}
+        self._functions = {}
+        self._prompts = {}
+
+    @property
+    def options(self) -> PromptManagerOptions:
+        """
+        Gets the configured prompt manager options.
+        """
+        return self._options
+
+    def add_data_source(self, data_source: DataSource) -> "PromptManager":
+        """
+        Registers a new data source with the prompt manager.
+
+        Args:
+            data_source (DataSource): Data source to add.
+
+        Returns:
+            PromptManager: The prompt manager for chaining.
+
+        Raises:
+            ApplicationError: If a data source with the same name already exists.
+        """
+        if data_source.name in self._data_sources:
+            raise ApplicationError(f"DataSource '{data_source.name}' already exists.")
+        self._data_sources[data_source.name] = data_source
+        return self
+
+    def get_data_source(self, name: str) -> DataSource:
+        """
+        Looks up a data source by name.
+
+        Args:
+            name (str): Name of the data source to lookup.
+
+        Returns:
+            DataSource: The data source.
+
+        Raises:
+            ApplicationError: If the data source is not found.
+        """
+        if name not in self._data_sources:
+            raise ApplicationError(f"DataSource '{name}' not found.")
+        return self._data_sources[name]
+
+    def has_data_source(self, name: str) -> bool:
+        """
+        Checks for the existence of a named data source.
+
+        Args:
+            name (str): Name of the data source to lookup.
+
+        Returns:
+            bool: True if the data source exists, False otherwise.
+        """
+        return name in self._data_sources
+
+    def add_function(self, name: str, function: PromptFunction) -> "PromptManager":
+        """
+        Registers a new prompt template function with the prompt manager.
+
+        Args:
+            name (str): Name of the function to add.
+            fn (PromptFunction): Function to add.
+
+        Returns:
+            PromptManager: The prompt manager for chaining.
+
+        Raises:
+            ApplicationError: If a function with the same name already exists.
+        """
+        if name in self._functions:
+            raise ApplicationError(f"Function '{name}' already exists.")
+        self._functions[name] = function
+        return self
+
+    def get_function(self, name: str) -> PromptFunction:
+        """
+        Looks up a prompt template function by name.
+
+        Args:
+            name (str): Name of the function to lookup.
+
+        Returns:
+            PromptFunction: The function.
+
+        Raises:
+            ApplicationError: If the function is not found.
+        """
+        if name not in self._functions:
+            raise ApplicationError(f"Function '{name}' not found.")
+        return self._functions[name]
+
+    def has_function(self, name: str) -> bool:
+        """
+        Checks for the existence of a named prompt template function.
+
+        Args:
+            name (str): Name of the function to lookup.
+
+        Returns:
+            bool: True if the function exists, False otherwise.
+        """
+        return name in self._functions
+
+    async def invoke_function(
+        self, name: str, context: TurnContext, memory: Memory, tokenizer: Tokenizer, args: List[str]
+    ) -> Any:
+        """
+        Invokes a prompt template function by name.
+
+        Args:
+            name (str): Name of the function to invoke.
+            context (TurnContext): Turn context for the current turn of conversation with the user.
+            memory (Memory): An interface for accessing state values.
+            tokenizer (Tokenizer): Tokenizer to use when rendering the prompt.
+            args (List[str]): Arguments to pass to the function.
+
+        Returns:
+            Any: Value returned by the function.
+        """
+        function = self.get_function(name)
+        return await function(context, memory, self, tokenizer, args)
+
+    def add_prompt(self, prompt: PromptTemplate) -> "PromptManager":
+        """
+        Registers a new prompt template with the prompt manager.
+
+        Args:
+            prompt (PromptTemplate): Prompt template to add.
+
+        Returns:
+            PromptManager: The prompt manager for chaining.
+
+        Raises:
+            ApplicationError: If a prompt with the same name already exists.
+        """
+        if prompt.name in self._prompts:
+            raise ApplicationError(
+                (
+                    "The PromptManager.add_prompt() method was called with a "
+                    f"previously registered prompt named '{prompt.name}'."
+                )
+            )
+
+        # Clone and cache prompt
+        self._prompts[prompt.name] = deepcopy(prompt)
+        return self
+
+    async def get_prompt(self, name) -> PromptTemplate:
+        """
+        Loads a named prompt template from the filesystem.
+
+        The template will be pre-parsed and cached for use when the template is rendered by name.
+
+        Any augmentations will also be added to the template.
+
+        Args:
+            name (str): Name of the prompt to load.
+
+        Returns:
+            PromptTemplate: The loaded and parsed prompt template.
+
+        Raises:
+            ApplicationError: If the prompt is not found or there is an error loading it.
+        """
+        if name not in self._prompts:
+            template_name = name
+
+            # Load template from disk
+            folder = os.path.join(self._options.prompts_folder, name)
+            config_file = os.path.join(folder, "config.json")
+            prompt_file = os.path.join(folder, "skprompt.txt")
+
+            # Load prompt config
+            try:
+                with open(config_file, "r", encoding="utf-8") as file:
+                    template_config = PromptTemplateConfig.from_dict(json.load(file))
+            except Exception as e:
+                raise ApplicationError(
+                    (
+                        "PromptManager.get_prompt(): an error occurred while loading "
+                        f"'{config_file}'. The file is either invalid or missing."
+                    )
+                ) from e
+
+            # Load prompt text
+            sections: List[PromptSection] = []
+            try:
+                with open(prompt_file, "r", encoding="utf-8") as file:
+                    prompt = file.read()
+                    sections.append(TemplateSection(prompt, self._options.role))
+            except Exception as e:
+                raise ApplicationError(
+                    (
+                        "PromptManager.get_prompt(): an error occurred while loading "
+                        f"'{prompt_file}'. The file is either invalid or missing."
+                    )
+                ) from e
+
+            # Migrate the templates config as needed
+            self._update_config(template_config)
+
+            # Group everything into a system message
+            sections = [GroupSection(sections, "system")]
+
+            # Include conversation history
+            # - The ConversationHistory section will use the remaining tokens from
+            #   max_input_tokens.
+            if template_config.completion.include_history:
+                sections.append(
+                    ConversationHistory(
+                        f"conversation.{template_name}_history",
+                        self._options.max_conversation_history_tokens,
+                    )
+                )
+
+            # Include user input
+            if template_config.completion.include_images:
+                sections.append(UserInputMessage(self._options.max_input_tokens))
+            elif template_config.completion.include_input:
+                sections.append(UserMessage("{{$temp.input}}", self._options.max_input_tokens))
+
+            template = PromptTemplate(template_name, Prompt(sections), template_config)
+
+            # Cache loaded template
+            self._prompts[name] = template
+
+        return self._prompts[name]
+
+    def has_prompt(self, name: str) -> bool:
+        """
+        Checks for the existence of a named prompt.
+
+        Args:
+            name (str): Name of the prompt to check.
+
+        Returns:
+            bool: True if the prompt exists, False otherwise.
+        """
+        if name not in self._prompts:
+            folder = os.path.join(self._options.prompts_folder, name)
+            prompt_file = os.path.join(folder, "skprompt.txt")
+
+            return Path(prompt_file).exists()
+        return True
+
+    def _update_config(self, template_config: PromptTemplateConfig):
+        # Migrate old schema
+        if template_config.schema == 1:
+            template_config.schema = 1.1
+            if (
+                template_config.default_backends is not None
+                and len(template_config.default_backends) > 0
+            ):
+                template_config.completion.model = template_config.default_backends[0]