mirror of
https://github.com/microsoft/semantic-kernel.git
synced 2026-10-12 02:19:06 +00:00
Evan Mattson
8381e50e8b
### Motivation and Context The current `FunctionCallBehavior` has allowed us to utilize auto function calling for OpenAI type models. As we proceed to support more AI connectors, that differ from OpenAI models, we need to be able to handle functions for all models that support function calling. <!-- Thank you for your contribution to the semantic-kernel repo! Please help reviewers and future users, providing the following information: 1. Why is this change required? 2. What problem does it solve? 3. What scenario does it contribute to? 4. If it fixes an open issue, please link to the issue here. --> ### Description In this PR: - We introduce a new function calling abstraction called `FunctionChoiceBehavior` which has three types: Auto, Required, and None. - One is able to configure the `function_choice_behavior` along with fully qualified function names (e.g., plugin1.function1), `maximum_auto_invoke_attempts` or `auto_invoke_kernel_functions` in yaml and JSON prompts. We have a new concept example showing how to do this for for yaml and JSON prompts. - If the fully qualified names are specified in the config file, they take precedence over the filters, if specified at a later point. - To make sure this isn't a breaking change, we still handle the previous `FunctionCallBehavior`, but so we can make decisions on the new `FunctionChoiceBehavior` we map the `FunctionCallBehavior` to `FunctionChoiceBehavior`. Each time `FunctionCallBehavior` is updated, `FunctionChoiceBehavior` will be updated, too. - The `_process_tool_call()` method in the `open_ai_chat_completion_base` needs to maintain the argument name as `function_call_behavior` as we cannot introduce a breaking change. - All concept samples have been converted to use `FunctionChoiceBehavior`. - New unit tests have been added for `FunctionChoiceBehavior` and we still currently support `FunctionCallBehavior` tests to make sure we haven't broken the backwards compatibility. - Added `deprecated` typing decorators to classes/methods to alert users that it would be best to transition to `FunctionChoiceBehavior` even though `FunctionCallBehavior` is still supported. - The `FunctionCallingStepwisePlanner` was updated to use the new `FunctionChoiceBehavior`. - Closes #6496, #6458 - Addresses #6626 <!-- Describe your changes, the overall approach, the underlying design. These notes will help understanding how your code works. Thanks! --> ### Contribution Checklist <!-- Before submitting this PR, please make sure: --> - [x] The code builds clean without any errors or warnings - [x] The PR follows the [SK Contribution Guidelines](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md) and the [pre-submission formatting script](https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md#development-scripts) raises no violations - [x] All unit tests pass, and I have added new tests where possible - [x] I didn't break anyone 😄
SK Python Documentation Examples
This project contains a collection of examples used in documentation on learn.microsoft.com.
Prerequisites
- Python 3.10 and above
- Install Semantic Kernel through PyPi:
pip install semantic-kernel
Configuring the sample
The samples can be configured with a .env file in the project which holds api keys and other secrets and configurations.
Make sure you have an Open AI API Key or Azure Open AI service key
Copy the .env.example file to a new file named .env. Then, copy those keys into the .env file:
GLOBAL_LLM_SERVICE="OpenAI" # Toggle between "OpenAI" or "AzureOpenAI"
OPEN_AI_CHAT_MODEL_ID="gpt-3.5-turbo-0125"
OPEN_AI_TEXT_MODEL_ID="gpt-3.5-turbo-instruct"
OPENAI_API_KEY=""
OPENAI_ORG_ID=""
AZURE_OPEN_AI_CHAT_DEPLOYMENT_NAME="gpt-35-turbo"
AZURE_OPEN_AI_TEXT_DEPLOYMENT_NAME="gpt-35-turbo-instruct"
AZURE_OPENAI_ENDPOINT=""
AZURE_OPENAI_API_KEY=""
AZURE_OPENAI_API_VERSION=""
Note: if running the examples with VSCode, it will look for your .env file at the main root of the repository.
Running the sample
To run the console application within Visual Studio Code, just hit F5.
Otherwise the sample can be run via the command line:
python.exe <absolute_path_to_sk_code>/python/samples/documentation_examples/planner.py