.Net Agents - No ability to maniplate history for KernelFunction based strategies #8914
Assignees
Labels
agents
enhancement
experimental
Associated with an experimental feature
.NET
Issue or Pull requests regarding .NET code
Comments
crickman
added
.NET
4 tasks
|
This is a pain point for me right now. Need to control what history is part of the executed prompt to reduce ambiguity (ie LLM has too much context, so will get confused), save tokens. Looking forward to this being resolved. |
github-merge-queue bot
pushed a commit
that referenced
this issue
Sep 20, 2024
…hat` (#8913) ### 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. --> Refine handling of history for `KernelFunctionSelectionStrategy` and `KernelFunctionTerminationStrategy` based on customer input. Fixes: #8898 Fixes: #8914 ``` Determine which participant takes the next turn in a conversation based on the the most recent participant. State only the name of the participant to take the next turn. No participant should take more than one turn in a row. Choose only from these participants: - ArtDirector - CopyWriter Always follow these rules when selecting the next participant: - After CopyWriter, it is ArtDirector's turn. - After ArtDirector, it is CopyWriter's turn. History: [ { "Role": "user", "Content": "concept: maps made out of egg cartons." }, { "Role": "Assistant", "Name": "CopyWriter", "Content": "Navigate your world, one carton at a time." } { "Role": "Assistant", "Name": "ArtDirector", "Content": "Approved. The copy effectively conveys the concept with a clever and concise tagline." } ] ``` ### Description <!-- Describe your changes, the overall approach, the underlying design. These notes will help understanding how your code works. Thanks! --> - Introduce convenience method (static) to create a `KernelFunction` from prompt-template and specifiy "safe" parameters. - Internally utilize `ChatMessageForPrompt` wrapper for `ChatMessageContent` to eliminate the inclusion of extraneous metadata in the strategy prompt. - Update `KernelFunctionSelectionStrategy` and `KernelFunctionTerminationStrategy` to call `ChatMessageForPrompt.Format` to format history for strategy prompt. - Add optional `HistoryReducer` property to `KernelFunctionSelectionStrategy` and `KernelFunctionTerminationStrategy` to limit how much history is included in the strategy prompt. - Updated sample ### 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 😄
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Customer Reported:
Always including the entire history can waste tokens, and unnecessarily increase operational cost.
In addition, the history may eventually exceed the token limit.
The text was updated successfully, but these errors were encountered: