discord.py /slash command modal for LLM config, improved API error handling. AI Discord bot, part 3

Resources:

[scarecrowb]

Yes, a tutorial on implementing langchain into the bot would be cool. It also has a tool for searching videos on youtube. And it could also serve as an intro for introducing agents in futre videos.

[hexa]

The way you handle the max 2k tokens limit for discord is to do fixed splits at 1.8k tokens, but that can result in the text being cut at weird points. Wouldn’t it be possible to use the streaming function in llama and post the messages to discord when there’s a carriage return? That way you get the text split by paragraphs (I doubt those go beyond 2k tokens) which are easier to read and look more “human-like”. As a side benefit it’d also help when using big models that go at like 4 tokens per second and would take too long to wait for 2k+ response to fully finish generating.

[trstmeimadctr]

Suggestion: Can you do a video on langchain’s Web Research Retriever and huggingface’s HuggingChat about how to implement web search to get up to date info when prompting? Both have public githubs. Particularly, huggingface claims to improve the answers with RAG and local embeddings but are a bit vague about how that plays a role in the process

[leminhquan]

Great tutorial sir! Hope you'll do a tutorial on how to make this bot fetch data from a database like mysql phpmyadmin database. Still this was a great video!