You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
By leverage web-llm project, we can run small LLM in web browser via WebGPU API.
This ticket is about creating a simple frontend chatbot leverages / show case Hybrid Search feature that's in development.
This will be an optional feature that can only be turned on when hybrid search is on.
Due to the limitation of running LLM in browser (mainly, small context window), we will not implement multi-step reasoning (for complex task execution). Instead, we will only implement simple tool calling pattern + context (which page are we) awareness. Tools available to agent will depends on the page the user is at.
On Home, Search pages, available tools:
Dataset search tool: can search catalog for relevant datasets based on the user's question
Greet user tool: Great user and tell user which feature is available based on current location
Dataset page: Agent should try look at if Tabular data analyst tool can serve user's question before use other tool
Dataset search tool
Greet user tool
Tabular data analyst tool: this tool will only be used if:
the dataset contains at least one tabular file
the tabular file is relevant to user question
Distribution page:
Dataset search tool
Greet user tool
Tabular data analyst tool: this tool will only be used if:
the distribution contains a tabular file
the tabular file is relevant to user question
The text was updated successfully, but these errors were encountered:
Simple Frontend Chatbot Leverages Hybrid Search
By leverage web-llm project, we can run small LLM in web browser via WebGPU API.
This ticket is about creating a simple frontend chatbot leverages / show case Hybrid Search feature that's in development.
This will be an optional feature that can only be turned on when hybrid search is on.
Due to the limitation of running LLM in browser (mainly, small context window), we will not implement multi-step reasoning (for complex task execution). Instead, we will only implement simple tool calling pattern + context (which page are we) awareness. Tools available to agent will depends on the page the user is at.
The text was updated successfully, but these errors were encountered: