Version 1.2
Release 1.2 supports all features of 1.1.
New features and improvements:
- Data encryption for all storage types.
- SharePoint list
- Database
- Local Storage
- Seamless support of the Chat Sharing option for all storage types
- Max chat name length is limited to 150 chars for encrypted storages and 255 for plain ones.
- GPT-4 Vision APIM endpoint (/openai4/vision).
- Prerequisites:
- To enable the deployment of Vision-preview in Azure Open AI, you must configure the Azure AI Computer Vision resource in the same Azure region as your GPT-4 with Vision resource.
- More information: https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/gpt-with-vision
- By default, this resource uses Price Tier: Free (F0), which supports 20 Calls per minute, 5K Calls per month.
- You can choose the more capable Price Tier: S1, which supports 10 Calls per second and costs ~1$ per 1000 calls.
- The recommended zone for the deployment is Swedish Central (as of Dec 2023). It supports both types of imaging models Vision and Dalle3 for image generations.
- To enable the deployment of Vision-preview in Azure Open AI, you must configure the Azure AI Computer Vision resource in the same Azure region as your GPT-4 with Vision resource.
- Optional checkbox to enable Vision model under the enabled integrations.
-
Speech synthesis to read out AI-generated texts by default
- The standard Web Speech API requires selecting the preferred language; using default page language is not always optimal. This default option is used when Native OpenAI text-to-speech model is not available.
- Azure OpenAI does not yet have support for text-to-speech models (as of Dec 2023).
-
/tts APIM endpoint (/openainative/tts) and Native OpenAI text-to-speech model for AI-generated texts.
- In case of using Native OpenAI endpoints - direct URL or APIM-based operation - the native text-to-speech model automatically handles text that contains mixed languages.
Bug fixes
a. Revised the PDF summarization logic to suppress content errors related to rapidly exhausted max content length limits in old models like GPT 3.5 Turbo 16k.
- The new logic does not send previously processed PDF content of the same chat when the new PDF document has been loaded for processing.
b. Optimized image compacting logic for images uploaded to be analyzed with GPT-4 Vision model.
- Uploaded images are saved to Chat Images if the library exists.
- Otherwise uploaded images are compressed to JPEG format (as in Version 1,1).