ChatGPT made my interview questions for me (Streamlit + LangChain)
Vložit
- čas přidán 23. 07. 2024
- Twitter: / gregkamradt
Newsletter: mail.gregkamradt.com/signup
Jupyter Notebook: github.com/gkamradt/langchain...
Streamlit Repo: github.com/gkamradt/llm-inter...
Streamlit App (I'll change it to Bring-Your-Own-API-Keys Soon): gkamradt-llm-interview-resear...
Full Streamlit Tutorial: • Build Your Own OpenAI ...
Workaround Otken Limits: • Workaround OpenAI's To...
LLM Assisted Interview Prep: Gather Research & Summarize People
0:00 - Intro
0:48 - Jupyter Code
9:58 - VS Code
12:00 - Streamlit
16:00 - Results
17:02 - Deploy
18:52 - Outro
A lot to digest! Thank you for breaking it down!
Wow, that was a well structured video. Seeing this the night before an interview, about to git clone and dive right in! Great material 🤌
This is absolutely epic!
Nice!! Thanks Connor
This was Great
big fan greg
so fire
amazing
hand-pink-waving
input_documents is not in the prompt in the notebook. Is that a typo?
It's a hidden parameter that LangChain uses under the hood. Not super clear I know
github.com/hwchase17/langchain/blob/8fdf88b8e3da9a5744b7a13afa99b16529438a31/langchain/chains/combine_documents/map_reduce.py#L187
Why did you need to make 2 prompts for this?
One prompt for the map step, one prompt for the combine step.
Since we had 3 chunks we needed to process
@@DataIndependent thanks. What exactly does the map part do? I just see that you generate some questions and then refine them in second prompt? Shouldn't the first prompt be a summarization prompt? Seems like extra steps for not much reward? Do you need to use map reduce for the context length being too long? I probably don't understand it fully. thanks!
Hi Greg, Big fan here. I've been watching your videos and tweets for a while now. It's been awesome and educative.
I want to ask, do you advise using Langchain and a backend framework like Django to build APIs that works with LLMs.Thank you!
Hey! LangChain works with a bunch, at the end of the day it's just an interface for a language model so it would work with any web framework you'd like. Django is great to keep the python going
This is fantastic, I know many sales people who do this manually before a meeting with a prospect. Being able to pull from LinkedIn and consuming 10ks are what they do regularly rather than twitter as a primary social media avenue. Unfortunately, LinkedIn makes it non-trivial to grab someones posts. Have you tried using LI versus Twitter (or in addition)?
I haven’t done a ton with LinkedIn due to their strict api policies.
But I’ve used phantom buster with good success there