Player not loading? Watch on YouTube
Uploaded on August 7, 2024, this tutorial demonstrates a local MindSearch installation using InternLM2.5-7b-chat. The speaker describes an Ubuntu 22.04 system with an NVIDIA RTX A6000 and 48 GB of VRAM. That is the demonstration hardware, not an established minimum requirement. The setup reflects the project at the time of upload; no MindSearch version or separate recording date is supplied.
The walkthrough creates an isolated environment, clones the repository and installs its requirements. The speaker starts the backend in English with a local model server, then launches a Gradio frontend in another terminal. He also describes a GPT-4 option that requires an OpenAI API key. He says the local model supports Chinese and English, with more optimization for Chinese.
The first query triggers a model download, and the interface reports an error while it is still downloading. An early search also retrieves information about someone with a similar name. After reloading and refining the query, the speaker gets a response and explains how the planner builds a graph of subquestions while the searcher retrieves web information.
A second example requests a Beijing-to-Western-Sydney trip itinerary. The demonstration covers flights, accommodation and daily activities. This is a self-hosted web search workflow that uses online sources. The speaker praises its results, but the examples do not establish answer accuracy or parity with Perplexity Pro.
