A browsing plugin for Huggingface based chat LLM models 🤗
👆📲 To use, click on the "Open in Colab" button above.
📂 Check out the notebooks directory for another copy.
🤖 Using the smallest open-sourced Dolly 2.0 model (3B) by default. I found it very capable with the browsing plugin.
🙅♀️ I was not able to try out other similar models because of GPU restrictions, but all Huggingface transformers compatible Chat LLM models should work. 😎
💻 There is a field in the notebook to set a model of your choice.
search_topic - The search topic. This is different from the user questions. This can be things like "Trappist-1 planets", "SVB collapse" etc.
show_search_results - When checked, prints out the parsed content that is passed to the LLM.
snippets_to_use - The number of search snippets to use for content when open_links is uncheked. The model seems to find the snippets very useful for answering questions.
open_links - Whether or not to click on search engine results and scrape content from the different pages.
links_to_open - The number of pages to click through for content when open_links is checked.
max_page_chars - Contents approximately greater then this number is truncated. Large numbers may not fit in the GPU, and might make LLM response slower.
use_wikipedia - Only uses Wikipedia for the search_topic when checked.
max_wiki_article_chars - Similar to max_page_chars, but for Wikipedia articles.
🤌 Screenshot of the crawler in action
🤌 Screenshot of some outputs
