mirror of
https://github.com/microsoft/semantic-kernel.git
synced 2026-10-12 02:19:06 +00:00
### 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. --> We previously found that agents might get carried away from in long contexts. The persona adaptation messages were designed to steer the agents in an orchestration to respond more closely to the instructions provided (as system message). As we perform more testing with newer models, we believe that the persona adaptation messages are no longer necessary. The additional instability (because they are hardcoded messages that devs can't override) and cost (the additional token cost which is not significant, but they do pollute the context which makes debugging difficult) they introduce is not worth the benefit they provide. Closing #12294 > Ideally, we'd want to approach this more analytically. The lack of a multi-agent evaluation process makes it difficult. All samples were verified to make sure they behave the same before and after the removal of the persona adaptation messages. ### Description <!-- Describe your changes, the overall approach, the underlying design. These notes will help understanding how your code works. Thanks! --> 1. Remove the persona adaptation messages. 2. Unit tests. 3. Replace `OpenAIChatCompletion` with `AzureChatCompletion` in some of the samples. ### 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 😄