microsoft/semantic-kernel
1
0
Fork 0
mirror of https://github.com/microsoft/semantic-kernel.git synced 2026-10-12 11:19:56 +00:00
Code Issues Packages Projects Releases Wiki Activity
1,167 Commits 64 Branches 1,077 Tags
3ce874bea3d5fc9847ac6f31fa7a963fa09babdd
Commit Graph
7 Commits
Author SHA1 Message Date
Mark KarleandAbby Harrison 2733a5574f Python: Import OpenAPI documents into the semantic kernel (#2297) 
### Motivation and Context

<!-- 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.
-->

This allows us to import OpenAPI documents, including ChatGPT plugins,
into the Semantic Kernel.

### Description

<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
- The interface reads the operationIds of the openapi spec into a skill:
```python
from semantic_kernel.connectors.openapi import register_openapi_skill
skill = register_openapi_skill(kernel=kernel, skill_name="test", openapi_document="url/or/path/to/openapi.yamlorjson")
skill['operationId'].invoke_async()
```
- Parse an OpenAPI document
- For each operation in the document, create a function that will
execute the operation
- Add all those operations to a skill in the kernel
- Modified `import_skill` to accept a dictionary of functions instead of
just class so that we can import dynamically created functions
- Created unit tests

TESTING:
I've been testing this with the following ChatGPT plugins:
- [Semantic Kernel Starter's Python Flask
plugin](https://github.com/microsoft/semantic-kernel-starters/tree/main/sk-python-flask)
- [ChatGPT's example retrieval
plugin](https://github.com/openai/chatgpt-retrieval-plugin/blob/main/docs/providers/azuresearch/setup.md)
- This one was annoying to setup. I didn't get the plugin functioning,
but I was able to send the right API requests
- Also, their openapi file was invalid. The "servers" attribute is
misindented
- [Google ChatGPT
plugin](https://github.com/Sogody/google-chatgpt-plugin)
- [Chat TODO plugin](https://github.com/lencx/chat-todo-plugin)
- This openapi file is also invalid. I checked with an online validator.
I had to remove"required" from the referenced request objects'
properties:
https://github.com/lencx/chat-todo-plugin/blob/main/openapi.yaml#L85

Then I used this python file to test the examples:

```python
import asyncio
import logging

import semantic_kernel as sk
from semantic_kernel import ContextVariables, Kernel
from semantic_kernel.connectors.ai.open_ai import AzureTextCompletion
from semantic_kernel.connectors.openapi.sk_openapi import register_openapi_skill

# Example usage
chatgpt_retrieval_plugin = {
    "openapi": # location of the plugin's openapi.yaml file,
    "payload": {
        "queries": [
            {
                "query": "string",
                "filter": {
                    "document_id": "string",
                    "source": "email",
                    "source_id": "string",
                    "author": "string",
                    "start_date": "string",
                    "end_date": "string",
                },
                "top_k": 3,
            }
        ]
    },
    "operation_id": "query_query_post",
}

sk_python_flask = {
    "openapi":  # location of the plugin's openapi.yaml file,
    "path_params": {"skill_name": "FunSkill", "function_name": "Joke"},
    "payload": {"input": "dinosaurs"},
    "operation_id": "executeFunction",
}
google_chatgpt_plugin = {
    "openapi":  # location of the plugin's openapi.yaml file,
    "query_params": {"q": "dinosaurs"},
    "operation_id": "searchGet",
}

todo_plugin_add = {
    "openapi":  # location of the plugin's openapi.yaml file,
    "path_params": {"username": "markkarle"},
    "payload": {"todo": "finish this"},
    "operation_id": "addTodo",
}

todo_plugin_get = {
    "openapi":  # location of the plugin's openapi.yaml file,
    "path_params": {"username": "markkarle"},
    "operation_id": "getTodos",
}

todo_plugin_delete = {
    "openapi":  # location of the plugin's openapi.yaml file,
    "path_params": {"username": "markkarle"},
    "payload": {"todo_idx": 0},
    "operation_id": "deleteTodo",
}


plugin = todo_plugin_get # set this to the plugin you want to try

logger = logging.getLogger(__name__)
logger.addHandler(logging.StreamHandler())
logger.setLevel(logging.DEBUG)

kernel = Kernel(log=logger)
deployment, api_key, endpoint = sk.azure_openai_settings_from_dot_env()
kernel.add_text_completion_service(
    "dv", AzureTextCompletion(deployment, endpoint, api_key)
)

skill = register_openapi_skill(
    kernel=kernel, skill_name="test", openapi_document=plugin["openapi"]
)
context_variables = ContextVariables(variables=plugin)
result = asyncio.run(
    skill[plugin["operation_id"]].invoke_async(variables=context_variables)
)
print(result)
```

### Contribution Checklist

<!-- Before submitting this PR, please make sure: -->

- [ ] The code builds clean without any errors or warnings
- [ ] 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
- [ ] All unit tests pass, and I have added new tests where possible
- [ ] I didn't break anyone 😄

---------

Co-authored-by: Abby Harrison <[email protected]>
 2023-08-04 17:48:21 +00:00
Sailesh RAbby HarrisonAbby HarrisonShawn Callegari
 		0d37271499 Python: Action Planner implementation (#1948) 
### Motivation and Context
<!-- 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.
-->
This PR is a first draft for implementation of ActionPlanner in Python.
I try to follow the C# implementation as much as possible.
Fixes: #1917

### Description
<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
- I have implemented ActionPlanner in the
`python/semantic_kernel/planning/action_planner` directory. The
implementation is in a file called `action_planner.py` and the prompt
used by the planner (a replica of the prompt used in C#) is found in
`skprompt.txt` in the same directory.
- The functions `good_examples`, `edge_case_examples` and
`list_of_functions` are used by the prompt to populate examples and list
of available functions in the kernel.
- After creating the plan, I parse the JSON generated by the planner by
using regex matching to extract out pure JSON as implemented in #1496
- Based on whether or not the planner was able to identify a function, I
create a Plan instance and return it.

### Here is some example code for using the planner
```python
import logging
import semantic_kernel as sk
from semantic_kernel.planning import ActionPlanner
from semantic_kernel.connectors.ai.open_ai import (
	OpenAIChatCompletion,
	OpenAITextCompletion
)
from semantic_kernel.core_skills import (
	MathSkill,
	FileIOSkill,
	TimeSkill,
	TextSkill
)

async def main():
	kernel = sk.Kernel()
	api_key, org_id = sk.openai_settings_from_dot_env()

	kernel.add_text_completion_service("dv", OpenAITextCompletion("text-davinci-003", api_key, org_id))
	kernel.import_skill(MathSkill(), "math")
	kernel.import_skill(FileIOSkill(), "fileIO")
	kernel.import_skill(TimeSkill(), "time")
	kernel.import_skill(TextSkill(), "text")

	planner = ActionPlanner(kernel)
	ask = "What is the sum of 110 and 990?"
	plan = await planner.create_plan_async(goal=ask)
	result = await plan.invoke_async()
	print(result.result)

if __name__ == "__main__":
	import asyncio
	asyncio.run(main())
```

### Queries
- In the C# implementation, when populating the function parameters in
the plan object, the `plan.parameters` variable is updated with the
function parameters. When I did the same in python, these parameters
weren't being passed onto the function in `invoke_async` method of the
plan instance. The parameters are only being passed when I update them
in the `plan.state` variable. This code is in `line 121` of
`action_planner.py`. What is the right way to do this?
- Some skills / plugins require one of the inputs to be passed as the
`input` variable and not in the context. How should this be handled in
the planner?

![image](https://github.com/microsoft/semantic-kernel/assets/69807105/967ea4b7-914e-46ac-89fb-1eb81c4ea9b0)
**OR**

![image](https://github.com/microsoft/semantic-kernel/assets/69807105/13a21556-8028-4f54-b40a-e4ebd1c5e654)

### Issues that I faced
- The `text-davinci-002` text-completion model regularly output
incorrectly formatted JSON with commas missing. For example
  ```json
  {""plan"":{
""rationale"": ""The list contains a function that allows for writing to
files, so the plan is to use that function to write the desired content
to the file 'output.txt'."",
  ""function"": ""fileIO.writeAsync"",
  ""parameters"": {
  ""input"": null,
  ""content"": ""A brief history of computer science""
  ""path"": ""output.txt""
  }}}
  ```


Pending Items
- [x] Create unit tests
- [x] Send the function parameters to the planner in list of available
functions with/without input parameter.

### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [x] The code builds clean without any errors or warnings
- [x] The PR follows SK Contribution Guidelines
(https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
- [x] The code follows the .NET coding conventions
(https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions)
verified with `dotnet format`
- [x] All unit tests pass, and I have added new tests where possible
- [x] I didn't break anyone 😄

---------

Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Shawn Callegari <[email protected]>
 2023-07-31 18:29:50 +00:00
Sailesh RAbby HarrisonAbby Harrison
 		a2da17d336 Python: google connector (#2071) 
### Motivation and Context

<!-- 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.
-->
Implementation of Google search connector for web search skill.

Fixes: #1862 

Sample usage can be found in
`python/samples/kernel-syntax-examples/google_search_skill.py`

### Description

<!-- 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#dev-scripts)
raises no violations
- [x] All unit tests pass, and I have added new tests where possible
- [x] I didn't break anyone 😄

---------

Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
 2023-07-28 19:56:22 +00:00
JoowonAbby HarrisonAbby HarrisonShawn Callegari
 		484b39d144 Python: Add sequential planner to python (#1916) 
### Motivation and Context
<!-- 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.
-->
fixes #1915 

### Description
<!-- 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: -->
- [ ] The code builds clean without any errors or warnings
- [ ] The PR follows SK Contribution Guidelines
(https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
- [ ] The code follows the .NET coding conventions
(https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions)
verified with `dotnet format`
- [ ] All unit tests pass, and I have added new tests where possible
- [ ] I didn't break anyone 😄

---------

Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Shawn Callegari <[email protected]>
 2023-07-28 00:08:31 +00:00
Sailesh RAbby HarrisonAbby Harrison
 		eab7a8f63a Python: implemented web search engine skill with bing connector (#1813) 
### Motivation and Context
<!-- 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.
-->
In this PR, I have tried my hand at an implementation of web search
engine skill in python semantic kernel using the Bing Web Search API.

### Description
<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
In the semantic kernel directory, I have added a new directory called
web_skills (To replicate Skills.Web from C#) and added the web search
skill here. For now, I have implemented web search using the bing web
search API. If this approach is fine, then I can implement the same with
the google search API too. I have tried to stick with similar naming
conventions as used in the C# implementation with matching context
parameters and arguments.

I can also add some unit tests for the connectors and the search skill,
and add something like exponential backoff to avoid rate limit errors
while querying the search APIs.

Here is some sample code that checks the working of the search skill.

```python
import os
import semantic_kernel as sk
from semantic_kernel.web_skills.web_search_engine_skill import WebSearchEngineSkill
from semantic_kernel.web_skills.connectors import BingConnector
from semantic_kernel.connectors.ai.open_ai import OpenAITextCompletion

async def main():
    kernel = sk.Kernel()
    api_key, org_id = sk.openai_settings_from_dot_env()
    kernel.add_text_completion_service(
        "dv", OpenAITextCompletion("text-davinci-003", api_key, org_id)
    )
    connector = BingConnector(api_key=os.getenv("BING_API_KEY"))
    web_skill = kernel.import_skill(WebSearchEngineSkill(connector), "WebSearch")

    prompt = "Who is Leonardo DiCaprio's current girlfriend?"
    search_async = web_skill["searchAsync"]
    result = await search_async.invoke_async(prompt)
    print(result)

    """
    Output:
    ["Celebrity Celebrity News Everything You Need to Know About Leonardo DiCaprio and Camila Morrone's Relationship From the beginning of their romance to today, we track their relationship here. By..."]
    """

    prompt = """
    Answer the question using only the data that is provided in the data section. Do not use any prior knowledge to answer the question.
    Data: {{WebSearch.SearchAsync "What is semantic kernel?"}}
    Question: What is semantic kernel?
    Answer:
    """

    qna = kernel.create_semantic_function(prompt, temperature=0.2)
    context = kernel.create_new_context()
    context["count"] = "10"
    context["offset"] = "0"
    result = await qna.invoke_async(context=context)
    print(result)

    """
    Output:
    Semantic Kernel is an open-source SDK that lets you easily combine AI services like OpenAI, Azure OpenAI, and Hugging Face with conventional programming languages like C# and Python. By doing so, you can create AI apps that combine the best of both worlds. Semantic Kernel is at the center of the copilot stack.
    """

if __name__ == "__main__":
    import asyncio
    asyncio.run(main())
```

### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [x] The code builds clean without any errors or warnings
- [x] The PR follows SK Contribution Guidelines
(https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
- [x] The code follows the .NET coding conventions
(https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions)
verified with `dotnet format`
- [ ] All unit tests pass, and I have added new tests where possible
- [x] I didn't break anyone 😄

---------

Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
 2023-07-17 17:54:19 +00:00
Sebastian SantoyosnehaAlisha MaddyAlisha MaddyAbby HarrisonSneha
 		378a1a949c Python: Add Logit-Bias to TextCompletion abstractions (#1880) 
### Motivation and Context
<!-- 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.
-->
1. Adds the Logit-Bias feature that is in the C# into Python.
2. It adds the token bias configuration to the text and chat completion
settings.
3. It contributes to flexibility in prompt engineering by allowing the
user to influence the likelihood of specified tokens appearing in the
text and chat completion.
4. Closes #1854 

### Description
<!-- Describe your changes, the overall approach, the underlying design.
These notes will help understanding how your code works. Thanks! -->
We added a dictionary field called token_selection_biases to the classes
CompleteRequestSettings and ChatRequestSettings to implement logit bias
and included an example to demonstrate functionality.
### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [x] The code builds clean without any errors or warnings
- [X] The PR follows SK Contribution Guidelines
(https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
- [x] The code follows the .NET coding conventions
(https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions)
verified with `dotnet format`
- [x] All unit tests pass, and I have added new tests where possible
- [X] I didn't break anyone 😄

---------

Co-authored-by: sneha <[email protected]>
Co-authored-by: Alisha Maddy <[email protected]>
Co-authored-by: Alisha Maddy <[email protected]>
Co-authored-by: Abby Harrison <[email protected]>
Co-authored-by: Sneha <[email protected]>
 2023-07-14 00:04:00 +00:00
Shawn Callegariandname 7a3fad9dbb Python: move kernel-syntax-examples to python/samples folder (#1650) 
Moving kernel-syntax-examples to the python/samples folder.
Leaving README behind to point to the new folders (will delete in a few
sprints)

### Contribution Checklist
<!-- Before submitting this PR, please make sure: -->
- [X] The code builds clean without any errors or warnings
- [X] The PR follows SK Contribution Guidelines
(https://github.com/microsoft/semantic-kernel/blob/main/CONTRIBUTING.md)
- [X] The code follows the .NET coding conventions
(https://learn.microsoft.com/dotnet/csharp/fundamentals/coding-style/coding-conventions)
verified with `dotnet format`
- [X] All unit tests pass, and I have added new tests where possible
- [X] I didn't break anyone 😄

Co-authored-by: name <email>
 2023-06-23 15:19:39 +00:00