v2.1.2
·
1110 commits
to main
since this release
Release Notes
Overview
This update brings important bug fixes for setting up Azure OpenAI Services, and improvements in the About page's support form.
Groups
New Features
Classobjects included the API key parameter in every call and response, and used to determine if a class had an API key set up before allowing assistants or threads to be set up. There is now a newclass/{classId}/api_key/checkendpoint for this purpose that returns a boolean. (#633)
Models
Resolved Issues
- Fixed: Setting up Azure API keys and creating an Azure OpenAI client might fail because of a misnamed parameter being passed in the
openai.AsyncAzureOpenAIcall. (#633) - Fixed: When trying to set an API key, the AI Provider is always set as openai, even when the user selected Azure as the AI Provider. (#633)
- Fixed: Available models in the Azure AI Services might fail to load because the Azure OAI Client model objects have a
created_atinstead of acreatedfield, and noowned_byfield. (#633) - Fixed: Available models in the Azure AI Services might fail to load because the Azure client
/listendpoint returns different available model names than the OAI client. (#633) - Fixed: Saving an assistant with an unavailable deployment name shows an unhelpful generic error generated by an unhandled
openai.NotFoundError. Now, users are shown an error with the required deployment name. (#633)
Known Issues
- PingPong cannot verify the capabilities of an Azure model deployment using the Azure client. Therefore, we rely on the user to create deployments that correspond to the model name. For example, PingPong will assume that deployment "gpt-4o-mini" corresponds to model "gpt-4o-mini" with Vision capabilities, but it can be the deployment of any OAI model.
Workaround: Make sure the model deployment names match the model your deployment is using. - Azure model deployment names need to exactly match the model name provided by the /list endpoint. PingPong will generate an error if the deployment name cannot be found in the Azure endpoint provided.
Workaround: Make sure the model deployment names match the model names provided by the/listendpoint. PingPong will let you know the exact model deployment name it expects to find.
Support Form
New Features
- Users are asked to confirm that they do not wish to provide their contact information, acknowledging that the PingPong team will not be able to contact them to resolve their issues. (#631)
- Feedback or support message categories are now required for sending a message. (#634)
Deployment Information
| Schema Upgrade | Migration Script | Permissions Update | Task Definition Update | Configuration Update |
|---|---|---|---|---|
| No | No | No | No | No |
Deployment Details
- N/A
Related PRs
- Azure API Deployment-related Bug Fixes and Improvements by @ekassos in #633
- Bump @sveltejs/kit from 2.5.26 to 2.8.3 in /web/pingpong in the npm_and_yarn group across 1 directory by @dependabot in #632
- About page: Confirm user does not want to include contact info by @ekassos in #634
Full Changelog: 502-srv148-web107...505-srv149-web110