I built a GPT Investment Banker using this 312 PAGE document

[@v0rtex-]

Could you please once do a tutorial where you're not using OpenAI's API key, but rather link it through your local LLM that runs on your machine (like LLaMA or vincuna)? Also, something that is quite hard to find is how to fine-tune the weights of open-source models directly on your machines with a handcrafted custom dataset. After you could do like a test "before and after" to check if it really incorporated those new informations.

[@NicholasRenotte]

You got it! Will probably do something around non-OpenAI models this week!

[@CARdibia]

@@NicholasRenotte Thank you so much

[@utkarsh5138]

Some models that are open source and also that can be used for commercial purposes for example MPT-7B , models for stability ai , dolly etc. would be appreciated.

[@eftehassan9969]

​@@NicholasRenotte thanks for working on viewers feedback.

[@maciejb486]

@@NicholasRenotte Ayoooo, would be great!

[@justinnine4940]

Financial analysis is not document search. Try the following question: Read though all the pages and ignore management commentary, find out the real performance of the bank comparing to previous year. If your AI can give you a decent answer it passes as an financial analyst.

[@Gos2]

can you please create a video on a model that can replace openai? preferably cheap/free option?

[@mcdaddy42069]

I really appreciate your fast paced and concise tutorials with langchain more please

[@adairjones11]

Gosh, a youtuber that actually responds to his viewers...so awesome. Your video came up on my Google discovery, so i watched. I've never seen your content before, but it does touch on the AI stuff i follow. I don't have a clue what you showed here but i watched every second, and will binge your other stuff. You opened another dimension for me. I have some learning to do for sure. Thank you and ditto on everyone who commented

[@NicholasRenotte]

I try my best 😅 thanks so much for checking it out Adair!!

[@MattRouletista]

I'm giving up on running this, keep getting errors when trying to run based on the readme instructions

[@ibunker8533]

Short and sweet to inform and entertain. Nice work !.

[@abdullahebrahim4045]

Question. How do you do this for all enterprise data. Which could mean terabytes of data as well as many different file type formats. And then allow that to use any open source, not restricted to open ai and connect to the internet. Could you make a video plz

[@whatitdo6287]

I have been using GPT4 for trade analysis to include measuring profitability, order fill statistics and high level market data as it relates to my system and it gets it wrong quite often not to mention the analysis approach is different each time. I've used the Noteable plugin and it's ok but I'm looking to templatize an approach using a data feed from my EC2 machines for dashboarding purposes.

[@reidgajewski6755]

Help! When installing ChromaDB with pip install chromadb i keep getting this error "clang: error: the clang compiler does not support '-march=native' error: command '/usr/bin/clang' failed with exit code 1 [end of output] note: This error originates from a subprocess, and is likely not a problem with pip. ERROR: Failed building wheel for hnswlib Failed to build hnswlib ERROR: Could not build wheels for hnswlib, which is required to install pyproject.toml-based projects" I've been trouble shooting this all day and cannot get it resolved. MacOS 13.4 Intel i7

[@nicotine_freedom]

Holy shit my man, this is incredible. Not sure if it’s worth it to you but would love to see a tutorial on getting chatgpt to pull actual from peer reviewed articles. Need to find claims for my company but the research for this is ENDLESS.

[@ozgurgulerx]

Hey Nicholas, good video. Realy like the visuals. When you are showing the progression of code 2:50 is this github copilot? It looks better than typing and is faster so seems better to use in a video.

[@NicholasRenotte]

Nope, it’s me screen recording!

[@ozgurgulerx]

@@NicholasRenotte ah ok, just copying pasting code? Curious how the code appears word by word rather than standard typing visual.

[@cnclubmember23]

Is there a way to add multiple pdf files? Or if we can add only one then approximately how many pages can there be in that pdf?

[@konradriedel4853]

Exactly what i was looking for, TY. Some questions/Problems though, you May Help? 1 what about Unicode PDFs Like German lang, i get several utf8 Error using pypdf loader 2 is the llm still able to answer General questions Not refering this document? 3 whats the Cost/number of tokens for that prompt - only prompt and answer or does the vectorizing of the Long doc also Count into the openai Key Tokens? Thanks a Lot, keep IT Up!!

[@NicholasRenotte]

1. Hmm, not too sure about this, just saw some people are using an alternate PDF service. Will need to dig into it. 2. Yes, it'll only refer to the toolkit (aka the PDF) when needed 3. I couldn't find on the OpenAI site whether input tokens count to the usage amount. But it was approaching the 4k limit when I tried some other prompts.

[@Soulioza]

@@NicholasRenotte 2. Apparently not, asking general questions (not linked to the pdf) results in `I don't know`.

[@arjunparmar6121]

this wont work incase of multiple pdf files of annualreport with tables in it. it will pickup random file information as both the file may have same table with same names.

[@jakekisiel7399]

Man, you are doing really good on making new videos, as they keep better all the time. And, this one was great. So, keep it up. I appreciate you. Thanks.

[@NicholasRenotte]

Thanks a mil Jake! I'm pushing the hell out of myself to make them better and better each time. They're taking longer to make but I feel like it's worth it!

[@mohitbarhate6970]

​@@NicholasRenotte Please I Need An Server Based Facial Recognition From Android By Using Android Camera Realtime if Possible Audio (mic and Speaker ) also For Ai

[@mohitbarhate6970]

​@@NicholasRenotteAs well As If possible Make Am Discord Channel For Easier Chat Directly To You From Subscribers

[@JonasSandstroem]

Correct me if I’m wrong, it is t this just using the vector search first method and then injecting the snippet into the GPT prompt? I don’t see any fine tuning here or the mode actually learning based on the document ? You’re still limited by the context window of the model, right, and you would t be able to make whole conclusions based on the full document all at once? Rather it would be several snippets that are used in the prompt, running the risk of missing nuances i between?

[@code___f1]

This is FANTASTIC! Thank you! One question I wanted to ask was how to adjust the code now that Davinci is depreciated?

[@Cypherthegreatest]

Hi Nicholas Could you possibly do a short video on longer outputs which would normally require the user to type continue , also any tips on improving output speed when using tools some of the stuff im testing almost takes 300 seconds before it finishes giving me a json response thanks your videos are awesome btw

[@NicholasRenotte]

Hmm, I didn’t even consider this but will do some digging around!

[@Cypherthegreatest]

@@NicholasRenotte That would be great thank you :D, Also a video on using mutiple tools at once would be amaazing thank you once more for your hard work.

[@yusufkemaldemir9393]

Great. Too fast though. Here is the question, how many tokens OpenAI is using in its model and what is the page limit of PDF that can handle? Other one is which hugging face repo would be closest to Open AI tokenization size? I have used “google flan xxl”, which I did not find its token size good and the answers correct? Thanks

[@NicholasRenotte]

1. Close to hitting the 4k limit with this app, when I had temperature higher in fact I did hit the limit. 2. Mosaic Storywriter is way bigger in terms of input tokens, I think it can hit 65kish

[@Drone256]

Can you recommend use cases for something like this?

[@lorisdeluca610]

Why not setting temperature to just 0 so that the LLM does not make up stuff?

[@NicholasRenotte]

My bad, I changed it in the sample code when I recorded the vid I still had it set to .9

[@reinerheiner1148]

I'd love to see content on using local llms with commercial license, like: how good do they work with tools? how much does their smaller context window hinder their performance compared to gpt4, all in the context of langchain, vertial dbs and tools. There are so many videos about those llm's but practially none on what I wrote as far as I know. are llms like wizard mega even viable to build apps around or are they just too bad compared with gpt4? Can they use tools? So many questions ;) Great content btw, you earned a sub!

[@РусланЗалевских]

Thanks for the great video! But I could only work with one page of the PDF ( If I load more, it says "RateLimitError: Rate limit reached for default-text-embedding-ada-002 in organization on tokens per min. Limit: 150000 / min. Current: 0 / min.. Does it depend on the rate in Open AI?

[@MrgoldenRose]

any answer to this???

[@andresgomezvillasanti]

@@MrgoldenRose Let me see if I understood (SORRY for the delay). But what I understand is that you did not do the partitioning, with that the page division part should have been solved.

[@jwia007]

❤ I want you to know that for someone who was really having a tough time getting it and feeling defeated. I really appreciate this video and your others. Langchain integration has stumped me for a while and now I have finally got it working!!! Big thanks 🎉🎉

[@NicholasRenotte]

Yesssss!! Go getem, I’ve got more videos planned in the series!

[@AllexxZang]

Love it! Could you help me understand why we need to build this bot rather than simply feeding ChatGPT that 300 pages of pdf and asking it questions? Budget saving? I'm new to this.

[@shaplist6815]

Bless you my Goodman, i hope to see more series like these.

[@NicholasRenotte]

Plenty more on the way!!

[@shaplist6815]

@@NicholasRenotte looking forward to it

[@kevinehsani3358]

You are one of few people I know that can do a great job of running a review and comparison on various vector DB such as pinecone and chroma db. Would you mind doing one please?

[@NicholasRenotte]

Yah, you got it. Want to do that after the free models video!

[@MicheleHjorleifsson]

2 questions / commments 1. if you wanted to store the vector database so you could call it up later (to speed up the interaction by not tokenizing every load) is that possible 2. could you do a modified version with mpt7 or one of the other huggingface models

[@zepposprojects3205]

This 1st one is what I've been wondering about too.

[@user-wr4yl7tx3w]

Given the questions asked, why do we need agents for this?

[@NicholasRenotte]

It's not acting on the document but it could interact with other sources by leveraging the PDF. I didn't show it here for the sake of brevity but could just as easily add in a Python toolkit and have it do something with the PDF.

[@luis96xd]

Wow, this was an amazing tutorial, everything was well explained 312 pages is a lot, and the app works so fast I would like to see another video without using OpenAI API key, for example, hugging face hosted models or other open-source models, please As you explain so well, so I subscribed to your Machine Learning Course 😄

[@NicholasRenotte]

Thank you so much Luis! Let me know how you go with it! Definitely already working on the next video as of this morning!

[@Danielbg9655]

Your videos are amazing and really informative. I would love to know more about the tabular data agents Langchain has and their connections with other tools to make further operations with the retrieved data (calculations, charts, etc.). Keep it up!

[@NicholasRenotte]

Ooooh I like this, will add it to the list Daniel!

[@prithvip5959]

Have to Train an NLP in order to automate the process of Creating Functional flows and Programming flows for mobile applications and Web applications from Text. I mean it has to include the decision nodes, Loops, Multiple paths etc etc. The decision nodes , paths and loops will be provided by us via text, where in it should be able to visualize the flow and draw the user flows and programming flows etc. I think this application will cut down massive amount of Human Intervention and Time in order to create new applications.

[@danbennett2232]

Great video mate! Loved it 👌

[@NicholasRenotte]

Cheers Dan!!

[@dragonwave2652]

Hi. Is it possible to finetune the model by feeding the book. I need information from the book, which is used to generate answer from the document I will give. In the book there are some rules which is used to get information from the document. So can I teach a book. So then I could filter information from my document?

[@kevinehsani3358]

Thank you for this tutorial. I ran it as it is and worked just fine. Then I substituted the pdf file with a resume pdf file and changed nothing else and when prompted by streamlit I asked if this resume is a good fit for a data scientist and the code crashed saying chromadb is not loaded!?

[@NicholasRenotte]

Weird, I wonder if it’s a memory problem. So it just dies when you swap them out?

[@kevinehsani3358]

@@NicholasRenotte If I run the prompt "how suitable is this resume for a data scientist" on your pdf returns I don't know. When I use my resume as a pdf and run the same prompt it errors like this "NameError: name 'chromadb' is not defined Traceback:" and the last line "file "C:\Users\kehsa\AppData\Local\Programs\Python\Python310\lib\site-packages\langchain\vectorstores\chroma.py", line 117, in __query_collection except chromadb.errors.NotEnoughElementsException:" Could it possibly be that the file is actually too small!!

[@atultiwari88]

Thank you very much for this awesome tutorial. I have one question though. How can we store the document's embedded vector db so that we don't have go through this step everytime. Thank you once again

[@NicholasRenotte]

Haven’t looked into it but I’d imagine you can preserve the state of the ChromaDB store and reload it.

[@atultiwari88]

@@NicholasRenotte Thank you for your reply. I have figured it out.

[@MrRojido]

@@atultiwari88 - what was the solution?

[@BEYONDPREMIERE]

Can you tell how much money it takes with the api?

[@NicholasRenotte]

Spent maybe $3.50 for all of my LangChain videos so far.

[@Mihaitafox]

Really like this new fast format. Can it learns to code based on specific documentation? Like interacting with certain blockchain RPCs.

[@NicholasRenotte]

NGL I was super nervous about the new style but I wanted to test it out. You could add a Python toolset as well and definitely have it do it!

[@peralser]

Great Video as usual!! Amazing way to share your knowledge! Thanks Nicholas.

[@NicholasRenotte]

Cheers Sergio!! Much appreciated!

[@MattRouletista]

I'm running into an error on Line 35 in regards to Altair. I have v5, so it gives the following error: ModuleNotFoundError: No module named 'altair.vegalite.v4'. If I install v4, I run into the following error: AttributeError: module 'altair.utils' has no attribute 'deprecation'. Stuck between a rock and a hard place, any help much appreciated!

[@irshaadbahemia666]

Hello Nicholas Renotte, do you make use of tools such 'itertools' and 'functool' a lot during data science projects building? What about people just starting to build their data science portfolio, do they need to master those tools before getting started?

[@NicholasRenotte]

Nah, skip them if you’re just starting out. I only use them when absolutely necessary which isn’t often.

[@irshaadbahemia666]

Thanks so much, Nicholas Renotte for responding to me given the fact that I am still trying to learn python tools as much as possible before building a data science portfolio project and maybe l must touch upon some mathematical topics also.

[@aniketbhunia7752]

What if I want to feed tabular data instead of pdf

[@uzersayed3530]

can you do video on video frame interpolation ?

[@NicholasRenotte]

Nothing planned on that space just yet, might do something a little later this year!

[@rudynausch8495]

Thanks for the demo, @Nicholas. Have you tried HyperDB at all? Would be interested to hear your thoughts vs. Chroma.

[@NicholasRenotte]

Not yet, only other vector db I've looked at is Pinecone so far! Will take a look soon though, might need to do a comparison video.

[@matt37221]

can someone explain how he gets the link running after finihsing code?? like what is that one step to get it running?

[@savageninja7710]

How would the token usage look like for a document this large? Also, what other open source models from huggingface or smt could i use instead of ones from openai?

[@NicholasRenotte]

Wasn’t too bad, haven’t spent more than a few dollars for all my OpenAI vids so far!

[@ittechzone8036]

Can you create a video to extract the life science-related data projects?

[@NicholasRenotte]

Got an example use case for me?

[@ittechzone8036]

@@NicholasRenotte yes. Could you help me create a video

[@HikiMaxSimo]

Great job! Great video... super great channel! It would be great as well if you could share which version of python you're using to be consistent with your code and not to have any issue with requirements Thx a lot!

[@teachmehowtoduggi]

hey Nick do you have the link for the paper? Love the tutorial btw

[@NicholasRenotte]

The document? I added it to the GH repo!

[@micbab-vg2mu]

Thank you for the great video. I will try to use it.

[@NicholasRenotte]

Awesome!! Let me know how you go!

[@Sir.Black.]

What would be the difference the create this and using Azure Cognitive Search instead? I'm not a coding expert but I see some benefits of keeping your data and responses private but maybe a bit more expensive with ACS?

[@NicholasRenotte]

Haven’t looked at ACS unfortunately but I do something similar at work for IBM. Keeping stuff private is critical for highly regulated companies and government!

[@amrutha331]

Quick question - Since we are using embeddings, how are tables getting converted to vectors?.. Do nested tables get read with the same accuracy?

[@mikew2883]

Good stuff! Keep it coming! 👍

[@ukaszKozio]

great work as always but pace is a bit too frantic to my taste :) not difficulty-wise which is similarly easy to adapt but your voice and presentation is already super energetic and interesting to watch and compressing it into a very short material with such a high and constant level of energy is just less pleasant and absorbing as longer materials with more natural pauses and changes of pace - contents are super helpful as always though, thanks

Hey, dude! I'm absolutely loving your video. The production quality is top-notch, and I adore how you showcased the vector database. Haha! Thanks a bunch for sharing it.

[@d_b_]

How does this work under the covers? For each prompt, is it first comparing key words of the prompt to key words of the document, then append the most relevant section to the prompt and finally submit?

[@ameerazam3269]

can you train/fine tune llm for custom model .. and used by langchain. ? if possible ...

[@angiee5271]

Hi are the responses that fast for you, or is the video sped up/cut? Because when i tried it on my 300 pages document, it took much longer to answer.

[@akshatapalsule2940]

Hello Nicholas getting error while installing chromadb can you suggest something else

[@CharlieIfrah]

Question: Why does it only work with that specific PDF? When I swap out the PDF for my own I get the following error: InvalidRequestError: This model's maximum context length is 4097 tokens, however you requested 6690 tokens (6434 in your prompt; 256 for the completion). Please reduce your prompt; or completion length.

[@ianfernandez7427]

could be your open ai api? im not confident but could be some sort of token limit on that side? i used a pdf smaller than the one he did and it didnt have errors

[@gregoryrutty2412]

Did you find a solution to this?

[@pritishghatkar6148]

Hi Nick is it possible to use Nvidia llm instead of Openai llm on the code that you have created?

[@NicholasRenotte]

Haven't seen an Nvidia LLM stuff outside of NeMo, going to be doing something on that soon!

[@adamgdev]

Do you know if this code will work well on a MacBook pro? I'm guessing it may run slower. I'm running into issues with all local gpt applications since I don't have an M1. Keen to hear your thoughts. Great video!!

[@NicholasRenotte]

It shouldn’t be too much slower the LLM processing is done remotely via the OpenAI API. I might do a short where I try to run it on my M1

[@somansh.2731]

if I create a custom chatbot with OpenAI API and feed in information that is highly confidential, do if i know if my information is secure?

[@fpena06]

Excellent tutorial! Could you please make one with vicuna or LLaMA. Thanks!

[@NicholasRenotte]

You got it Francis, should be out this week!

[@marwinsteiner5817]

Hiya, what if I need to convert my file(s) from .pdf to .tex first, do I need to use OCR to do that?

[@yazanrisheh5127]

I keep getting this error despite creating a brand new OpenAI account. I'm trying to create a chatbot that can read from a pdf file. How do I fix this error: "You exceeded your current quota, please check your plan and billing details" There's no way I can exceed if I literally just created my account...

[@chrismachabee3128]

Super video. You're a pretty smart guy. thanks for sharing. i love annual reports.

[@muralidhar2060]

Hi Nick...Great content and I subscribed! Is that possible to use an open-source model instead of OpenAI? Also, can you create similar to query Pandas data frames as well?

[@NicholasRenotte]

Sure is, didn’t have pandas planned for the next vid but I’ll sneak it in!!

[@mohamedaly1023]

Would this be possible without the OpenAI API? Using a local bin file LLM ?

[@thequestion3953]

This is actually insane. Had no idea LLMs were this flexible and portable. Thanks for uploading!

[@peepdaslan984]

I need help building a machine learning model. I don't know any code...

[@NicholasRenotte]

Take a look at some of the beginner tutorials I've got on the channel!! Lmk how you go!

[@FCrobot]

That is so brilliant video! thank you very much! But how can I move these tools online, please?

[@ahmednaguib6653]

Could you please help? I used the same code repo but on different PDF file and sometimes i get this error "This model's maximum context length is 4097 tokens, however you requested 4756 tokens (4500 in your prompt; 256 for the completion). Please reduce your prompt; or completion length." although the question and answer is very short. What about be he reason and how to fix?

[@CharlieIfrah]

Hi, I'm having the same issue. Did you find a solution?

[@GMCvancouver]

Is there any solution that can read Excel file privately without using open ai ?

[@RD-ej2cx]

Awesome video, really interesting use case! Is anyone else running into this error on langchain and chromadb module integration: "You must provide embeddings or a function to compute them"? Can't seem to debug it!

[@DockterWho]

Can you do one on llama cpp for local llms and also the hugging face hosted models? This was a great video!

[@eiyrm]

as it was written in chat gpt that don't send any personal data! is it good practice or ethical to use openai api in business and real world projects ? what are your thoughts about it

[@davidoak9510]

It's using OpenAI. ChatGPT is built on OpenAi but it isn't the same thing

[@nathandscott1]

An interesting experiment but doesn't really work. I ran it as is and it gave pretty random answers, you could see how it got them but they weren't accurate at all. "what was the impact of covid" and it tells you they supported the community with donations (which is nice) but mostly the pdf comments on financial negative impact. I then tried it on another PDF, and it was all over the place in its answers. It was a weekly progress report on projects. Maybe it would work on a product catalog or a doc with a clear structure. Anyway a cool little video and you do an excellent job on it, though this one was a little rushed, I liked the langchain one you did that was really well paced.

[@venkatakrishnanramesh4718]

I don't have open ai key, subscription edition. Can u pls guide. Me without it

[@NicholasRenotte]

Vid coming this week!!

[@vishwanathasardeshpande7092]

Can you create a video on using langchain used to develop chatbot which has a personality of a celebrity

[@semosemo3827]

if I have 1.5 million docx file each for a court record (between 1 to 100 page each)and I want to make chatbot on them, which tech you suggest me to work on?

[@danielit5845]

I cant install chromadb please help me, i dont Know what requirements i need to install. Im using Windows

[@tekkemganesh9424]

Hey hi can you create a model on confidence detection of a person

[@sonny8085]

Is this video available in English?...I tried turning on the CCs, but nothing changed 🙃

[@NicholasRenotte]

Hey David, refresh browser and clear cache? I just tried I got the auto gen subs.

[@g-program-it]

you just made something investment banks would pay thousands of £/$ for. Great job dude.

[@NicholasRenotte]

I'm off to speak to Goldman Sachs on Monday 😉, thanks for checking it out!

[@toddai2721]

I can see you becoming a billion air by 2035... all you have to do is... shhgfg... shuffle u trfbj... jbgdfbj.

[@AJvanuw]

I think the key is making something that’s not so easily reproducible

[@jpaulmeyer]

@@NicholasRenotte lol actually? Can make an intro

[@zamap4278]

​@@AJvanuwDefinitely, you'd wanna make the ingestion process more streamlined for end users and adapt some features for specific use cases but it's a great base to build off of

[@divaxshah9424]

I love your videos, but can you please create some where you don't use openai APIs , instead use the models like gpt4all-j or other which are open source, it would be really helpful, thanks for making this kinds of videos 😊

[@NicholasRenotte]

Coming this week 🎉

[@noobking5056]

9 days passed

[@NicholasRenotte]

@@noobking5056 my bad, still working on it. Rest assured it’s coming, I just kinda went nuts on it.

[@noobking5056]

@@NicholasRenotte no worries man I thought you had never reply, man you so popular, so I was working on langchain with one of the commercial open source AI models to train it on multiple school policies it would work right?

[@SumedhKadoo]

@@NicholasRenotte Subscribed, thanks

[@clivejefferies]

I might be missing something, but this looks more like a document information extractor/summary creator than an investment banker. Great tutorial nonetheless and I love the use of container to illustrate memory.

[@NicholasRenotte]

Thanks a mil Clive! This was a first pass at the task, will probs tighten it up and focus it moreso on a sales and trading env!

[@Franx570]

What about token consumption?

[@ashleigh3021]

Brilliant. People on WallStreetOasis have been saying for years this would never happen and that i-bankers were irreplaceable lmao.

[@datawithrajadwr1597]

unable to install chromadb, how you guys solved it

[@prashanthbs7698]

I followed your instructions and ran the program. On first question itself I got You exceeded your quota, please check your plan and billing details

[@MattRouletista]

Finally got this to work! Well worth the struggle

[@MattRouletista]

Is there any way to make the responses faster?

[@AnkitYadav-nm1uy]

Great. Please make one with Falcon 40B

[@picious]

are you publishing private content to OpenAI ?

[@NicholasRenotte]

Nah, this is all public!

[@mayien-0]

Nice vidéo, but i don't understand how you embbeding the PDF document to store in ChromDB

[@mayien-0]

Ok, I have my answer : it's chromaDB which embedding automatically with Sentence-Transformers

[@_rd_kocaman]

you can load pdf document to chromaDB by calling the function from_document() and it will create a embedding collection

[@UN5KO]

The Camera is not open.. Can yuo help me?

[@Mcbingus]

Are you using gpt4 api? also how do I change the document so its my own? cheers!

[@NicholasRenotte]

I don’t think the default is 4 when using the OpenAI service, I’ll look into it 🤔

[@criscanto7040]

Hey man your videos are the best! Not only knowledge based but also the format! Quick question, how can we define 3.5 turbo as the model to be used here? Bests!@@NicholasRenotte

[@Diego0wnz]

do you need openai pay for this?

[@NicholasRenotte]

Yep, but could use free models as well.

[@DJPapzin]

Great video, thank you for this tutorial.