GSoC Weekly Update: Refining the Model Pipeline and Designing the First Version of Chat Activity
Model Pipeline Adjustments
This week, I made adjustments to the temperature and the maximum token length for model-generated contents, and simplified the prompts provided to the model. You can view the latest generation results in the sheet.
In general, the quality of the model’s responses has improved. However, occasionally when calling the same model frequently, it generates a lot of repetitive content, as shown below.
UI/UX Design for Chat Activity
This week, I completed the UI/UX design for the first version of Chat Activity integrated with the Chatbot. Currently, I aim to make modifications based on the existing UI of Chat Activity without making major changes to the original interface. The following image shows the current running interface of Chat Activity.
The following image shows the introduction of the chatbot functionality in Private/Neighborhood Mode. When the user enters text, if they need to interact with the chatbot based on their previous input, they can click the newly added icon in the lower left corner to get the chatbot’s response. The chatbot will generate the response in a chatbox style, where ‘chatbox’ is a class/component currently used within the Chat Activity.
I also want to introduce a new mode for Chat Activity: Chatbot Mode. In this mode, users will interact only with the chatbot and not with other friends or users. The way it operates will be similar to many chatbot platforms we commonly use.
In Chatbot Mode, after the user inputs a message, the chatbot will directly process the user’s information and generate a response to communicate with the user. While the chatbot is generating content, the input box below will be disabled until the chatbot has completed generating the content.
For the Figma design proposal, you can check here. Looking forward to your suggestion and contribution!