Is your feature request related to a problem? Please describe.
Currently title generation is based only on the user's initial input prompt. While this works for most cases, it isn't as effective when the prompt requires the model to use a tool (e.g. webscraping) in order to obtain a more full context to answer the prompt. An example use case would be summarizing a webpage - in this case the model hallucinates the contents of the webpage when generating a title, often resulting in a totally off-topic title.
Describe the solution you'd like
Provide the context/output of any tool calls resulting from the initial prompt before generating the title. This will result in better, more informed conversation titles.
Describe alternatives you've considered
Another option is to wait till the model has sent an initial reply in the conversation, like what OpenAI does in their ChatGPT interface, but this is largely unnecessary for most conversations without tool use, and just including tool outputs should get us most of the way there in terms of correctness, and would maintain current performance for non-tool conversations.
Originally created by @ekatiyar on GitHub (Aug 9, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/4507
**Is your feature request related to a problem? Please describe.**
Currently title generation is based only on the user's initial input prompt. While this works for most cases, it isn't as effective when the prompt requires the model to use a tool (e.g. webscraping) in order to obtain a more full context to answer the prompt. An example use case would be summarizing a webpage - in this case the model hallucinates the contents of the webpage when generating a title, often resulting in a totally off-topic title.
**Describe the solution you'd like**
Provide the context/output of any tool calls resulting from the initial prompt before generating the title. This will result in better, more informed conversation titles.
**Describe alternatives you've considered**
Another option is to wait till the model has sent an initial reply in the conversation, like what OpenAI does in their ChatGPT interface, but this is largely unnecessary for most conversations without tool use, and just including tool outputs should get us most of the way there in terms of correctness, and would maintain current performance for non-tool conversations.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @ekatiyar on GitHub (Aug 9, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/4507
Is your feature request related to a problem? Please describe.
Currently title generation is based only on the user's initial input prompt. While this works for most cases, it isn't as effective when the prompt requires the model to use a tool (e.g. webscraping) in order to obtain a more full context to answer the prompt. An example use case would be summarizing a webpage - in this case the model hallucinates the contents of the webpage when generating a title, often resulting in a totally off-topic title.
Describe the solution you'd like
Provide the context/output of any tool calls resulting from the initial prompt before generating the title. This will result in better, more informed conversation titles.
Describe alternatives you've considered
Another option is to wait till the model has sent an initial reply in the conversation, like what OpenAI does in their ChatGPT interface, but this is largely unnecessary for most conversations without tool use, and just including tool outputs should get us most of the way there in terms of correctness, and would maintain current performance for non-tool conversations.