Development with Large Language Models Tutorial - OpenAI, Langchain, Agents, Chroma

Welcome to this course about development with Large Language Models, or LLMs. Throughout this course, you will complete hands-on projects will help you learn how to harness LLMs for your own projects. You will build projects with LLMs that will enable you to create dynamic interfaces, interact with vast amounts of text data, and even empower LLMs with the capability to browse the internet for research papers.
This course was developed by @Luup.ai123
Colab notebook for introduction to the API:colab.research.google.com/dri...
Github : github.com/pythonontheplane12...
Join the Discord: / discord
Twitter : / akshatnon
⭐️ Contents ⭐️
⌨️ (0:00:00) Brief introduction to LLMs
⌨️ (0:11:49) Quick note from the future
⌨️ (0:12:04) Chatgpt playground (skip this is you know this already)
⌨️ (0:18:21) GPT API basics (skip this is you know this already)
⌨️ (0:30:43) Brief intro to chainlit
⌨️ (0:31:33) Cloning chatgpt user interface
⌨️ (0:45:37) Limitations of our interface
⌨️ (0:47:48) Adding streaming, backend view, stop sequence button
⌨️ (0:58:42) Introduction to vector databases
⌨️ (1:04:42) Vector databases hands on
⌨️ (1:12:10) QnA with Documents - .txt and .pdf
⌨️ (1:23:32) Testing out our Q&A system
⌨️ (1:27:22) Introduction to web-browsing and agents
⌨️ (1:32:52) AI researcher
⌨️ (1:42:23) Human as a tool
⌨️ (1:44:44) Mini code interpreter plugin(Replit tool)
⌨️ (1:46:29) Searching youtube using agents
⌨️ (1:49:19) Guide to explore more
⌨️ (1:50:33) Shell Tool
⌨️ (1:55:43) Create your own tools
⌨️ (2:01:19) Ending Notes
🎉 Thanks to our Champion and Sponsor supporters:
👾 davthecoder
👾 jedi-or-sith
👾 南宮千影
👾 Agustín Kussrow
👾 Nattira Maneerat
👾 Heather Wcislo
👾 Serhiy Kalinets
👾 Justin Hual
👾 Otis Morgan
--
Learn to code for free and get a developer job: www.freecodecamp.org
Read hundreds of articles on programming: freecodecamp.org/news

[@Luup.ai123]

Hey everyone, Akshath here Had a great time making it this course and I hope you enjoy it 🎊🎊

[@sriramchundi3142]

Amazing video. Most simply explained video on the channel. Great work!!!

[@KararaJawaab]

Thank you so much. God bless you.

[@VictorRage]

thank you for dedicating your time to this course

[@leandroblanmo]

That's such an amazing content, Akshath. Is there any suggestion on how to have a deeper understanding of each step we went through in this video? That's a totally new field for me. Can you share a roadmap to master this subject? I mean, if I study Machine Learning only, will I learn everything from this lesson in depth? Many thanks!

[@AdeAulya]

This is great. Thank you for this explanation. i'll try in google collab if i may.

[@Bobcoolyoung]

This is exactly what I was waiting for.

[@jeroenritmeijer7283]

it appears that various libraries have moved on a bit since this was recorded. If you get errors during execution then make sure you replace 'message' with 'message.content' when passing that parameter into a function.

[@xu-raytjhin7315]

Thank you !

[@FatchelPiresGuitar]

Amazing, guys! I'll love a similar video using opensource models!

[@nithinbhandari3075]

Completed this video. Learned chainlit, document q and a by basics of chromadb (most favorite part), basics of vector db, langchain, agents etc. It was nice. The second hour was more interesting. (As first hour was basic stuff that many people may knows).

[@AIWithShrey]

I'm unsure why we used Chatlit instead of Streamlit, but it works nonetheless. Something very fun to do is to upload your resume and ask the chatbot questions on it. I think this will help the hiring team of an organisation greatly, as it simplifies the job of reviewing resumes and understanding the applicant's abilities. So scaling the document QnA project will be tremendously valuable.

[@Luup.ai123]

hi yes, thats a fair point as streamlit is more popular. I wanted this series to emphasize giving people an understanding of LLMs and their capabilities so i just chose chainlit because it needed less code to work with lets us only worry about the use of LLMs and agents.

[@DANNYEL20122]

@@Luup.ai123 hi please how can i run chainlit using google colab or jupyter notebook?

[@mayankkumawat6077]

My Godddd, the ability to have custom tools for the agent changes everything, so theoretically (I bet, we are going to see some implementations of these pretty soon), we can do everything using natural language, what a time to be alive

[@Alice8000]

This is exactly what I was waiting for. Before anyone else.

[@navturn]

Amazing man, I am hooked !

[@rewanthnayak2972]

Thank you for answering my request so quickly

[@markceraldi8719]

QUESTION: at VERY end of video, the custom tool multiplier decided to return "12" even when the delegated python function had hard-cod overrided to divide "/" instead of "*". The system internally experiences cognitive dissonance "thoughts" -- including the correct answer - then resolves by returning an unqualified 12 due to all the prevailing context suggesting that was the intention, eg some "common sense". I would love to know more about that clash like if there is a term for it. Also this was the most mind expanding tutorial ive ever seen in my life.

[@GattuBattu789]

Very Informative one

[@som4971]

finally i can understand LLMs thanks fcc

[@user-hy2el1pt1r]

Amazing video!

[@MariodeFelipe]

Good content and fantastic explanation 🎉

[@krox477]

Wow we already started using LLMs

[@louisejin792]

very very helpful ，tks

[@peeturpain9379]

Thank you for this. I have been assigned a project at our company for document QnA with LLMs and this is exactly what I wanted. Please upload follow up courses as well such as for prompt engineering.

[@letssee6813]

Hii @peeturpain9379. i have some doubts regarding your llm project.

[@peeturpain9379]

@@letssee6813 what do you mean?

[@Yogic-ignition]

HI i would love to talk about this, even i have a similar project going on

[@peeturpain9379]

@@Yogic-ignition sure, lets get in touch

[@tjhorton]

Correction: Think of parameters as WEIGHTS (between neurons) not as NEURONS.

[@Galmion]

when the trainer includes the API key in the main file lmao

[@RedCloudServices]

thanks for the video lesson. the new OpenAI retrieval Assistant abstracts all of these steps so was wondering if Chainlit can be configured with the new Assistant API?

[@akshayugale5785]

Hi @Luup.ai123 , Nice vide. Can you also make a video on RLHF or share some resource on it ? Like how to introduce this concept to your model.

[@conradcaldeira7131]

heyyy there, have always enjoyed your vids......could you provide a tutorial for creating a LLM from scratch on windows without GPU

[@vibuvignesh7359]

I am getting an error 13 permission denied error. How do I fix it?

[@bidhanchandrosana]

Good

[@SYEDHASSANASLAMBE-CSE]

Sir please post about tts model vedio

[@shivanshugautam1381]

Hi, I am working with document QA project but I am bit confused about what will be the actual chunk size for better semantic search results. And I am using milvus to store my embeddings. I am having large pdfs.

[@abdulrehmanbaber2104]

try different chunk sizes, they are made only because the context window of llms are short.

[@Yogic-ignition]

i am facing permission denied error after uploading the file. On the Temp folder. Any idea how to solve this

[@powerball200]

Do i need to know any prior skills to learn langchain? Like what coding language, web dev etc?

[@Robo-fg3pq]

When should we use langchain tools vs openai function calling?

[@user-jz6ks5wp2c]

can you help me for i build simulink model for this Dataset "Activity Human by smartphone"

[@csowm5je]

Thanks for tutorial. All developers moved to mac? 1:25 windows machine errno13 permission denied. got it to work with wsl.

[@nicholaslim2913]

I keep getting the error No module named 'chromadb'

[@manh9105]

don't see GPT-4 option in the free version of playground model? has open ai team removed it ?

[@Luup.ai123]

I dont think GPT-4 is available to non premium users. Bit of an oversight on my part using it as it as chatgpt premium is definetly not necessary for this course. The API is available to non premium subscribers as well as it works on a standard per use pricing model.

[@manh9105]

few questions Akshat. Who owns the data trained by us. If its Open AI ? aren't we making it more powerful basically like a Frankenstein

[@kingslyroche]

API and Playground requests will not be used to train our models.

[@Luup.ai123]

Hi, Akshath here. We dont specifically "train" it, we just do a retreival augmented search so technically chroma db is where all of this information is stored. Given that chroma is open source and the python package runs on our machinne locally, noone has access to the embeddings but us. Additionally OpenAI claims that it doesn't use any of our data to train their models. big emphasis on the word "claims".

[@user-qk2ld3zm4r]

Are we supposed to buy or pay credits for access to chatgpt playground?

[@c_54_spoorthi21]

can you make a tutorial to do the same with llama 2 model

[@muhammadaamir321]

Kindly upload prompt engineering course

[@rashedulkabir6227]

Prompt Engineering is not coding.

[@furkantyagi3261]

please made a video on tekla structure

[@KennethMak]

From buildspace 👋

[@wilonweb]

We have to get a GPT4 prenium account to do that ?

[@yaroslav8717]

Does it still make sense to go through his video after Assistants API release?

[@AB-kq9xm]

no

[@amansaxena4827]

👍

[@WHAT-gm1xm]

Learning hacking course actually can hack anything in technical field

[@kamalkamals]

Hope to make the same with langchain js not just langchain python

[@wilonweb]

You do that ?

[@Ruthrap-dm1ki]

this coding not support for windows is it?? if any another alternative code is available , let me know

[@amalvijay8222]

Rate limit exceeded for API key.. someone help me with this

[@ilianos]

If you've used gpt-4: there's a limit of requests per time allowed You can have a look into your settings when you log into your openai account. There you might have set a manual limit.

[@agastyareddy5489]

Ily

[@shubhrai3027]

can someone tell me if there's any free alternative of the OpenAi API ?......bcoz no GPT 4...and even GPT-3 is limited time , credit ....please.....thanks

[@41cata]

Does anyone have similar videos that use C++ or a different language instead?

[@MMABeijing]

python is easier than C++ ( a language that is not used in data science or AI). why would you need C++, if I may ask?

[@Abraham_doestech]

@@MMABeijing python may be easier than C++ to pick up but it is used in AI too. Especially for faster computations, ie: things like Pytorch and other Cuda directives have code in c++ not just in c as in the case of CPython etc etc. But yes not many people here will be using anything other than Python because of the ease of implementation

[@MMABeijing]

@@Abraham_doestech First I want to say that there is a C++ library to interact with Openai. Second: although I know the advantages of CPP, or rust, I am curious as to why @Majesticaln would want to use them to interact with OpenAi "infrastructure". There are fields where web assembly deemed mission critical, and considering that RUst actually has the tighest integration with WASM it makes sense to use it... but what is the use case that would motivate Majestic to want to talk with OpenAI using CPP? Python has a richer community, more help available, and langchain is using it too. It s clear to me that we are at a stage when the speed of integration is crucial, and for that reason I thing going along with python is the way to go. That being said I am interested in hearing other, possibly conflicting, opinions

[@duongital]

sorry for not relating much to the topic, what is the chrome extension for pets running?

[@abelgalindo1458]

Can someone provide a possible solution for this error? if match := _special_token_regex(disallowed_special).search(text): TypeError: expected string or buffer

[@digitalthinkers7760]

Issue: 1st param type is expected to be string or buffer in acall function. Resolution: So, you need to update the param to message.content. res = await chain.acall(message.content, callbacks=[cb]).

[@dovatmotos5223]

hello everyone! PermissionError: [Errno 13] Permission denied: 'C:\\Users\\Usuario\\AppData\\Local\\Temp\\tmpjf7ckm9u' can somone tell me how to fix this problem when i import the pdf to the app?! thanks!

[@raphaelmarietan9322]

Replace `with tempfile.NamedTemporaryFile() as tempfile` with `with tempfile.NamedTemporaryFile(delete=False) as tempfile`, since you don't want the temporary processed file you create to be deleted once closed.

[@nicholaslim2913]

Thank you!@@raphaelmarietan9322

[@salimnegan7289]

thank's, got same problem@@raphaelmarietan9322

[@jpw2184]

This should be pinned@@raphaelmarietan9322

[@gainmuscle]

javascript version anytime soon?

[@LordPBA]

19:15 why there is a cat walking on the screen?

[@bonkers_dave]

immediately stuck at playground UI, which does not look anything like the video, and coming into this cold, I have to go find some other resource or training to understand the playground UI.

[@yaroslav8717]

just change to Chat from Assistants next to Playground logo

[@user-rk4ux3cj8q]

1:41:27

[@ricardoandreom]

Do we need a chat gpt paid subscription to follow this tutorial?

[@KGnanaVarshita]

Do we need chat gpt subscription to complete this tutorial

[@Luup.ai123]

nope, you only need an api key

[@midprogramming]

Is there any reason why you used assistant over system when giving a personality?

[@techtechweekly]

18:30

[@NarutoHugsMikasa]

First !

[@41cata]

day1 - 34:44 day2 -

[@yaroslav8717]

Hard to follow the course, and not really a detailed explanation. But anyway thank you!

[@legosupermangustavo]

1st 😅

[@donelkingii3738]

Can you make it without the ethical blockers? Tired of the "That us illegal" blah blah etc.

[@shreyasaravanan7720]

100th comment

[@entertainmentarc6466]

Pappu , pedda pappu, panikiraani pappu

[@arsagor8506]

First comment plz pin

[@MyrLin8]

Aligning with human values ... not.

[@_MrCode]

These are very basics, and only 10 minute of content, but this guy takes 2 hours for extra talking, the main part of chatbot is multi-user conversation ability, which he didn't discussed even.