Recently, iPollo.ai has completed the system upgrade and the new version V1.2.0 is now online. The upgrade includes the following changes:
1.iPollo.ai has supported scheduling multiple large models.
It includes Microsoft Azure OpenAI Service, Replicate (supports Llama-2–70b-chat), HuggingFace (currently supports google/flan-t5-xxl and Deci/DeciCoder-1b), and more models will be added soon, including IFlytek Spark, ERNIE Bot, MINIMAX, etc. Users can select the appropriate large model for their application in the fine-tuning training interface.
2.Plugin system is officially launched.
Users can develop strategies for the application by configuring prompts and corresponding plugins, enabling the application to have networking, real-time data retrieval, and expert large model capabilities. The application can schedule plugin capabilities and respond to users’ questions at appropriate times based on prompt content. (Currently, plugin function is only supported for GPT-3.5 and Microsoft Azure large models. Other models will be opened gradually, so please stay tuned.)
3.Text-based application is officially launched.
Users can train designated roles in the background. After the training is completed, the application can generate text content of specified types according to user requirements.
4.Speech-to-text function is now available.
Users can enable the speech-to-text function for dialogue-based applications by clicking on the “Add Function” button in the fine-tuning training section. Once the function is activated, users can click on the microphone icon in the app chat interface to complete voice input.
5.API documentation has been published.
After creating an application, users can view the corresponding API documentation in the API section. Users can generate an application-specific API Key in the upper right corner of the page and request application interfaces according to the requirements in the API documentation.
6.Comprehensive upgrade of vector database.
After this update, the dataset supports the secondary editing of vector blocks. Users can perform secondary editing on the text content and keywords of any vector block after splitting the vector block in the Embedding model. This function can greatly increase the accuracy when reading the dataset and increase the flexibility when users create datasets. At the same time, users can also add new vector blocks and configure the text content and keywords in the vector blocks according to their needs.
7.Account application channel is now open. A new account application channel has been added to the iPollo.ai registration page. New users can fill out the form and check their application email within 24 hours to obtain their account password.
In addition to the upgrades of the important sections mentioned above, we have also fixed some bugs in the original version and completed some page upgrades. Users can directly apply for training their own AI on the official website according to their needs and data, for project consultation, customer service, community management, etc.
We have been expecting you!
All Comments