👏👏 Hey Liam, your five-minute tutorial is fantastic! Kudos and thanks for putting the effort to produce it. Your app is exactly what any knowledge worker is craving for: We all have gigabytes of pdf files in some folder named "READ", "TO READ" or "__TO READ" (so it stays on top of the root :), but never get to it (probably distracted by all these tutorials to become more productive we love to watch). A bot that can read that stuff for us, so we can continue to wing it is a true godsend. :D
Thank you it worked perfectly despite generating an error on the pip install. ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. yfinance 0.2.18 requires beautifulsoup4>=4.11.1, but you have beautifulsoup4 4.8.2 which is incompatible.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. yfinance 0.2.18 requires beautifulsoup4>=4.11.1, but you have beautifulsoup4 4.8.2 which is incompatible.
Hi Liam, I am getting 'authentication Error' when running 2. section of the code "Embed text and store embeddings" . I have not change anything yet just running it as is. Any suggestion?
It’s convenient because I just completed a Data Analysis course via IBM, and Vanderbilt Promp Engineering course. I created my first Smart Bot for my Dad’s website on Sunday. I’d like to dump RFP contractor documents to easily take the 88 pages to question parts of a bid
couple of things observed. 1. Its not free because integration with openAI is required 2. It is too slow. For two page PDF it takes somewhere around 10-20 seconds to respond when I am on a 48GPU machine
Is there any way I can test it for free? I used a PDF with only one page and it says "You exceeded your current quota, please check your plan and billing details"
The Best tool for this is th-cam.com/video/bcK7LldB3dk/w-d-xo.html I like some of the transitions, but sometimes they're a bit too much and are seemingly random. Since we use these persistent elements that transition across pages to indicate some kind of relationship between the previous and the next states, some of your transitions confuse me because I can't immediately see what the relationship is. For example 1:23 of the selectable tiles (which weren't selected) transition into being two switches... does that mean anything? are they related in some way? I see this as random and a bad use of the design language. However, at 3:14 I like the transition from switches to the ticks on a paper, that makes sense to me. Epic presentation tho
@@ahmedmiftah8308what’s the point in merging the pdfs if the chunks is going to break them up anyways Each section should be able to be another pdf which makes since anyways
This is great. Is it possible to retrieve images from the PDF? I have a PDF with many graphics that help understand the content. Do you have any ideas as to how I can provide images as part of the conversation?
For anyone runnning into rate limiting issues, I had success using an exponential backoff on the request. I replaced the entire first part of step 2 with the following: # Get embedding model from tenacity import ( retry, stop_after_attempt, wait_random_exponential, ) # for exponential backoff @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6)) def completion_with_backoff(): return OpenAIEmbeddings() embeddings = completion_with_backoff()
Thanks, but it would be interesting to do without the OpenAI API, as it is paid and it would be very expensive for large projects to be analyzing PDFs with it, it could be another hugging face model, I am trying to do something in this direction, if you have any ideas let us know let me know!
Liam, this is a great tutorial, thank you. What I really liked was the explanation of what is happening behind the scenes - anyone (even a non-developer) like me - can cut and paste the code but knowing what the commands are doing is super helpful. The explanations in the Colab are great and I took your advice and stole your code. The chatbot was up and running in a few hours (remember: non-developer) but that included building a separate UI. Great work, thank you
can i ask whether you paod for the OPENAI KEY OR YOU DID IT WITH THE FREE TRAIL? Cuz am encountering this error RateLimitError: You exceeded your current quota, please check your plan and billing details.
WARNING:langchain.embeddings.openai:Retrying langchain.embeddings.openai.embed_with_retry.._embed_with_retry in 4.0 seconds as it raised RateLimitError: You exceeded your current quota, please check your plan and billing details.. This pops up while I'm creating the vector db
Liam, if there an option to make the assistant always use the data that has been uploaded to knowledgebase? It doesn't read the KB files every time and uses the links that even doesn't exist
Out of complete ignorance, is Langchaining the best method currently available to increase the perform of our LLMs Chatbots? If not, what is it or what other methods are out there that I may be missing. Thanks for answering.
Hey Liam, How many PDFs can I use this on? I have 1000+ instructional documents on an information system I use and have been trying to create a chatbot with this database embedded for quick question answering. Would i have to combine all the pdfs? can i put them all through vectorization? What are your thoughts?
there is a little error, in the embedding section OpenAIembeddings is not defined, if I am not wrong just add a line -> import openai (I wonder on 115.760 views how many has really done the tutorial XD )
Thank you very much for this great video!!! One question. On the part of Create chat bot with chat memory (OPTIONAL), I received the following message "DeprecationWarning: on_submit is deprecated. Instead, set the .continuous_update attribute to False and observe the value changing with: mywidget.observe(callback, 'value'). input_box.on_submit(on_submit)" Why? Would you be able to fix it?
What if we have multiple PDfs and we want to fetch the Answer from that pdf ? like for an example : I have 20 Pdfs, and if I ask one question then it should fetch the answer from any one of the Pdf (correct obviously) and show me as a output.
great tutorial! I have hundreds of research papers in pdf format. Can I use this approach to build a vector db and then chat with chatgpt? Is there a limit to the size of db? any pitfall to avoid?thanks!
Thank you for your excellent sharing. This is great guidance, and I hope you can continue to share more! If there's anything I can do, please let me know~
i there! As a fellow filmmaker, I find the concept of regenerative agents fascinating. I'm curious, what specific types of agents are you interested in exploring in your video? Additionally, have you thought about incorporating some real-world examples of sim city-like models, such as the ones developed by Stanford, to help illustrate the concept to your audience? Looking forward to hearing more about your project! George Anton
Great video! I was wondering why is it a private chatbot when you're using openAI key and sending the information to LLM GPT-3.5? How can you secure sensitive data with your method? Thank you sharing your knowledge.
Hi Liam, great video. I do have a question, from the following code, i notice that we don't have to specifically turn the "query" into embeddings, before it performs a search against the vector db? Is it because the function "similiary_serach" internally calls the openapi embedding to perform words embeddings? query = "Who created transformers?" docs = db.similarity_search(query)
I am creating a chatbot to help employees, I have a 220 page contract pdf that I need my chatbot to be able to answer questions about accurately. The issue is, fine-tuning with the data doesn’t produce accurate outputs. Would this be a good way to achieve this?
So essentially you calculate semantic similarity of the stored vectors and the asked question, then provide the 4 most similar vectors as context in the prompt?
can i ask whether you paod for the OPENAI KEY OR YOU DID IT WITH THE FREE TRIAL? Cuz am encountering this error RateLimitError: You exceeded your current quota, please check your plan and billing details.
Thanks a lot man, been trying to get this to work via other ways for days. This was so easy, great tutorial. How would you transfer something like this to a user friendly ux/ui?
Not necessarily, if you cram it full of thousands of chunks I'd assume the recall just gets slower and slower and uses more resources on your system. Best to setup different indexes for different information or use namespaces (Pinecone feature)
Thank you, I've learned a lot from your channel. I'm curious about the differences between the llama index and the langchain. Maybe I'm still a beginner in AI and don't quite understand.
Thanks for the super video. I have a question: in the overview you show that ChatGPT3.5 is used, or that the query is last processed by 3.5. But in the code I can't find any reference to it. Where is my mistake?
Hi, THANK YOU for sharing your knowledge. Could please let me know how many PDF can we train using this technique and does this LLM remember what PDFs it has been trained on or do we have to train the LLM at before running the query?
Liam your content is unreal Some of the best I've seen so far This is hard knowledge You are brilliant What do you mean by 512 tokens on every chunk? Characters? I'll be waiting for a detailed masterclass Vicente
NameError Traceback (most recent call last) in () 1 # Get embedding model ----> 2 embeddings = OpenAIEmbeddings() 3 4 # Create vector database 5 db = FAISS.from_documents(chunks, embeddings) NameError: name 'OpenAIEmbeddings' is not defined
Problem with this approach I find out that Your Bot gonna reply on irrelvant question "Ask who is spiderman". So even after providing a prompt which clearly Instruct system Not to response If current context doesn't hold knowledge LLM gonna reply. How to handle this scenario?
Hello. Thanks for a great video. But i have financial statement pdf file and it contains tables in it. How can i achieve besy results out of it? Any suggestions or help would be more helpful. Thanks😊
![ตีสิบเดย์ [FULL] | อาจารย์โอเล่ ญาณสัมผัส ผ่าดวงคนดัง ประเทศไทยจะเป็นยังไงหลังจากนี้!](http://i.ytimg.com/vi/gszT2O76fL4/mqdefault.jpg)
Leave your questions below! 😎
📚 My Free Skool Community: bit.ly/3uRIRB3
🤝 Work With Me: www.morningside.ai/
📈 My AI Agency Accelerator: bit.ly/3wxLubP
Golden! Clear, concise info and a notebook! If it's too fast for some viewers, I'll remind that they can always show down the replay speed.
👏👏 Hey Liam, your five-minute tutorial is fantastic! Kudos and thanks for putting the effort to produce it. Your app is exactly what any knowledge worker is craving for: We all have gigabytes of pdf files in some folder named "READ", "TO READ" or "__TO READ" (so it stays on top of the root :), but never get to it (probably distracted by all these tutorials to become more productive we love to watch). A bot that can read that stuff for us, so we can continue to wing it is a true godsend. :D
This was definitely one of your better videos. You explained Langchain well and I’m glad you used the colab notebook instead of Jupyter or repl.
Thank you it worked perfectly despite generating an error on the pip install.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
yfinance 0.2.18 requires beautifulsoup4>=4.11.1, but you have beautifulsoup4 4.8.2 which is incompatible.
ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
yfinance 0.2.18 requires beautifulsoup4>=4.11.1, but you have beautifulsoup4 4.8.2 which is incompatible.
Thought it would be just another video on the subject, but you summarize in an awesome way! Great vid! Congrats
thank you for time, effort and generosity,
I wish very good things for you.
Excellent! Thank you for your hard work to put these together.
My pleasure! Thanks for watching
This video was copied from the beginning to the end from the channel Prompt Engineering
@@AlbyTheMovieCreator Oh wow I totally didn't know that. Thanks for the heads up! SMH😒
Can you feed it multiple pdf at the same time like a group of 300 or would you have to run each line individually.
Brother can you make video on how to use autogpt for beginners 😊
A beginner's question, if everything happens locally, why do we need an OpenAI API Key?
Hi Liam, I am getting 'authentication Error' when running 2. section of the code "Embed text and store embeddings" . I have not change anything yet just running it as is. Any suggestion?
It’s convenient because I just completed a Data Analysis course via IBM, and Vanderbilt Promp Engineering course. I created my first Smart Bot for my Dad’s website on Sunday.
I’d like to dump RFP contractor documents to easily take the 88 pages to question parts of a bid
couple of things observed.
1. Its not free because integration with openAI is required
2. It is too slow. For two page PDF it takes somewhere around 10-20 seconds to respond when I am on a 48GPU machine
Is there any way I can test it for free? I used a PDF with only one page and it says "You exceeded your current quota, please check your plan and billing details"
@@shanamin1561 You can not its entire smartness is dependent of open AIs embeddings and its not free.
The Best tool for this is th-cam.com/video/bcK7LldB3dk/w-d-xo.html
I like some of the transitions, but sometimes they're a bit too much and are seemingly random. Since we use these persistent elements that transition across pages to indicate some kind of relationship between the previous and the next states, some of your transitions confuse me because I can't immediately see what the relationship is.
For example 1:23 of the selectable tiles (which weren't selected) transition into being two switches... does that mean anything? are they related in some way? I see this as random and a bad use of the design language. However, at 3:14 I like the transition from switches to the ticks on a paper, that makes sense to me. Epic presentation tho
Hey Liam! Awesome...could you do one that scrapes data from blog/website for embedded chatbot for a blog?
Can I do it with multiple PDFs? like for each pdf I'll just chunk every page?
Merge pdfs
@@ahmedmiftah8308what’s the point in merging the pdfs if the chunks is going to break them up anyways
Each section should be able to be another pdf which makes since anyways
Wonderful tutorial. Thank you!
No worries 🤙🏼
This is great. Is it possible to retrieve images from the PDF? I have a PDF with many graphics that help understand the content. Do you have any ideas as to how I can provide images as part of the conversation?
What type of graphics are you talking about?
You could leverage Lang Chain's agent feature set to use computer vision to analyze your images.
@@MCroppered the PDF I have has images embedded and I was wondering if how I could recall these during a conversation
@@quinnherden I will research this. Thanks!
@@GiovaDuarte if you figure out how to bring images along with the conversation, let us know how please
For anyone runnning into rate limiting issues, I had success using an exponential backoff on the request. I replaced the entire first part of step 2 with the following:
# Get embedding model
from tenacity import (
retry,
stop_after_attempt,
wait_random_exponential,
) # for exponential backoff
@retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6))
def completion_with_backoff():
return OpenAIEmbeddings()
embeddings = completion_with_backoff()
Thanks for the great video! One question: which OpenAI model is used to retrieve the answer? Is it gpt-35-turbo or ada or...? Where is it defined?
Thanks, but it would be interesting to do without the OpenAI API, as it is paid and it would be very expensive for large projects to be analyzing PDFs with it, it could be another hugging face model, I am trying to do something in this direction, if you have any ideas let us know let me know!
Yeah i am also trying to build something like that without using the OpenAI API.
Me too
Do you found something interesting ? For large pdfs project the price is very important
Anyone found something?
Liam, this is a great tutorial, thank you. What I really liked was the explanation of what is happening behind the scenes - anyone (even a non-developer) like me - can cut and paste the code but knowing what the commands are doing is super helpful.
The explanations in the Colab are great and I took your advice and stole your code. The chatbot was up and running in a few hours (remember: non-developer) but that included building a separate UI. Great work, thank you
can i ask whether you paod for the OPENAI KEY OR YOU DID IT WITH THE FREE TRAIL? Cuz am encountering this error RateLimitError: You exceeded your current quota, please check your plan and billing details.
@@aradinac I used the paid for openAi key
@@AndrewSheves which one did you buy?
@@aradinac I have the same error because I have the account not paid, if you found another solution, pls let us know
Amazing Video. I have a question: Can Your notebook (code) run with muti file pdf?
Can i do it on jupyter notebook rather then using colab
WARNING:langchain.embeddings.openai:Retrying langchain.embeddings.openai.embed_with_retry.._embed_with_retry in 4.0 seconds as it raised RateLimitError: You exceeded your current quota, please check your plan and billing details..
This pops up while I'm creating the vector db
I am experiencing the same problem. Can you help me out.
Thanks Liam ... neat and fast as always; could you post another similar video doing the same thing with Llama index pls. I thought that was easier.
Liam, if there an option to make the assistant always use the data that has been uploaded to knowledgebase? It doesn't read the KB files every time and uses the links that even doesn't exist
What is a good way to split text in a textbook pdf because on one page it has 2 columns, text on the left and right side?
Hey Liam @ 03.22 you said we can upload pdf data by entering the pdf name. But what if we have more pdf, life for example I have 5 pdf?
I would love to see a video which helps me to deploy such a chatbot (created on colab) on a webpage.
Cheers, this is a brilliant video. Looking forward to making a bespoke AI.
Thanks for the awesome video Liam, do you need a GPU to do such a thing? it looks like a GPT license/subscription is required as well ?
Out of complete ignorance, is Langchaining the best method currently available to increase the perform of our LLMs Chatbots?
If not, what is it or what other methods are out there that I may be missing.
Thanks for answering.
Hey Liam, How many PDFs can I use this on? I have 1000+ instructional documents on an information system I use and have been trying to create a chatbot with this database embedded for quick question answering. Would i have to combine all the pdfs? can i put them all through vectorization? What are your thoughts?
Getting lots of qs like this, will make a video on chatting over many PDFs this week 👍🏻
@@LiamOttley It would also be cool if it could handle other files types such as epub , doc and markdown files.
Liam is there a way to use langchain with open source models such as Vicuna?
@@LiamOttley Just subbed and turned on notifications!! Love that your responsive and videos are great man!
@@LiamOttley waiting your next video.
great tutorial , can it be modified to support multiple pdfs ?
there is a little error, in the embedding section OpenAIembeddings is not defined, if I am not wrong just add a line -> import openai
(I wonder on 115.760 views how many has really done the tutorial XD )
Thank you very much for this great video!!! One question. On the part of Create chat bot with chat memory (OPTIONAL), I received the following message "DeprecationWarning: on_submit is deprecated. Instead, set the .continuous_update attribute to False and observe the value changing with: mywidget.observe(callback, 'value').
input_box.on_submit(on_submit)" Why? Would you be able to fix it?
I am having same issue
Is the optional segment should be on same py program?
great video!, it is possible to add more than 1 pdf with that code ?, will be possible to provide a code for multiple pdf ? thank you
I cant wait until we can expand this to all documents. I assume that is what Microsoft 365 Copilot will do.
Will you be sharing your Marcus Aurelius database u created previously? I was really looking forward to that
will need a video on how to do this for multiple pdfs
What if we have multiple PDfs and we want to fetch the Answer from that pdf ?
like for an example : I have 20 Pdfs, and if I ask one question then it should fetch the answer from any one of the Pdf (correct obviously) and show me as a output.
great tutorial! I have hundreds of research papers in pdf format. Can I use this approach to build a vector db and then chat with chatgpt? Is there a limit to the size of db? any pitfall to avoid?thanks!
Thank you for your excellent sharing. This is great guidance, and I hope you can continue to share more! If there's anything I can do, please let me know~
i there! As a fellow filmmaker, I find the concept of regenerative agents fascinating. I'm curious, what specific types of agents are you interested in exploring in your video? Additionally, have you thought about incorporating some real-world examples of sim city-like models, such as the ones developed by Stanford, to help illustrate the concept to your audience? Looking forward to hearing more about your project! George Anton
I'm facing this issue: module 'openai' has no attribute 'error'
Yo I’ve made plugins but don’t know how to test it so can you give some ideas .(I don’t have access to the plugins yet.I’m in the waitlist)
Thats a fantastic video and to the point and thanks for the code as well
Great video! I was wondering why is it a private chatbot when you're using openAI key and sending the information to LLM GPT-3.5? How can you secure sensitive data with your method? Thank you sharing your knowledge.
Using and hosting by yourself a custom open source LLM like llama or mistral
Hi Liam, great video. I do have a question, from the following code, i notice that we don't have to specifically turn the "query" into embeddings, before it performs a search against the vector db? Is it because the function "similiary_serach" internally calls the openapi embedding to perform words embeddings?
query = "Who created transformers?"
docs = db.similarity_search(query)
How can I implement this chatbot with my custom Db with a longterm memory and access from the phone ?? Any guide?
This would be very interesting
Thanks, very good content. Just a question to understand the market better: did I misinterpret your hourly rate at $997/45 mins?
Appreciate your hustle bro
I am creating a chatbot to help employees, I have a 220 page contract pdf that I need my chatbot to be able to answer questions about accurately. The issue is, fine-tuning with the data doesn’t produce accurate outputs. Would this be a good way to achieve this?
So essentially you calculate semantic similarity of the stored vectors and the asked question, then provide the 4 most similar vectors as context in the prompt?
You're awesome, Liam !!
Can you suggest alternatives to OpenAI in terms of embeddings and llm? They are too expensive their APIs
can i ask whether you paod for the OPENAI KEY OR YOU DID IT WITH THE FREE TRIAL? Cuz am encountering this error RateLimitError: You exceeded your current quota, please check your plan and billing details.
Were you able to get around the error ?
Thanks a lot man, been trying to get this to work via other ways for days. This was so easy, great tutorial. How would you transfer something like this to a user friendly ux/ui?
Ask ChatGPT4
Awesome tutorial. Cheers Liam
Is there a limit to the number of PDF chunks you can add to the vector DB?
Not necessarily, if you cram it full of thousands of chunks I'd assume the recall just gets slower and slower and uses more resources on your system. Best to setup different indexes for different information or use namespaces (Pinecone feature)
Those biceps too! 💪
Great Work! Can we do this with a local or a smaller language model ?
Thank you, keep going.
Thank you, I've learned a lot from your channel. I'm curious about the differences between the llama index and the langchain. Maybe I'm still a beginner in AI and don't quite understand.
Ask ChatGPT4
Great job... will run this on my writings/ book collection and my code snippets, and build an awesome, MeKnowledgeBase 😎
This is great Liam, thank you for sharing, what's the simple automated way to deploy this code to a basic online application/chat page
Straightforward and concise! Great explanation.
How do you extract the exact page number where the answer was found?
This is amazing! Can you teach us mindai?
Thanks for the super video. I have a question: in the overview you show that ChatGPT3.5 is used, or that the query is last processed by 3.5. But in the code I can't find any reference to it. Where is my mistake?
The default LLM for Langchains "OpenAI()" is text-davinci-003 and "ChatOpenAI()" is gpt-3.5-turbo I believe
Freaking Great Content! Keep Rocking 💯
Hi,
THANK YOU for sharing your knowledge. Could please let me know how many PDF can we train using this technique and does this LLM remember what PDFs it has been trained on or do we have to train the LLM at before running the query?
As a beginner coding their first ever plug-in, how long would it take to develop a high quality plug-in?
Liam your content is unreal
Some of the best I've seen so far
This is hard knowledge
You are brilliant
What do you mean by 512 tokens on every chunk? Characters?
I'll be waiting for a detailed masterclass
Vicente
Many people may have privacy issues, can we do this without open ai api? where document privacy will be maintained?
I appreciate the tutorial.. the lack of dark mode is total eye torture
It is possible to create this knowledge based with local llm model like vicuna, llma etc? ... hmm
Good question! I know that LangChain integrates with gpt4all, but so far I haven't found any videos or articles talking about it.
Great video! Would be cool to create a video similar with Apify and LangChain.
great video! help me to complete me knowlege about best praticies in prompt!
Awesome work
Can you use .txt files instead of PDFs? Great video and content. Thanks
Good question
Here’s how to do it with text files:
th-cam.com/video/c0YDDSWr3t0/w-d-xo.html
NameError Traceback (most recent call last)
in ()
1 # Get embedding model
----> 2 embeddings = OpenAIEmbeddings()
3
4 # Create vector database
5 db = FAISS.from_documents(chunks, embeddings)
NameError: name 'OpenAIEmbeddings' is not defined
Can you explain how we could use other llms than openai, for example can we use mosaic mpt-7b ?
Anybody running in to this installation issues: DEPRECATION: textract 1.6.5 has a non-standard dependency specifier extract-msg
Problem with this approach I find out that Your Bot gonna reply on irrelvant question "Ask who is spiderman". So even after providing a prompt which clearly Instruct system Not to response If current context doesn't hold knowledge LLM gonna reply. How to handle this scenario?
i am noob here. Is it possible to embed it on a site? If I embed, is it standalone? or still it uses GPT API calls and costs?
Can you share video on voice based search in pdf Document.
Where does he describe the model to use for output? Is he using Da Vinci 003?
Thanks
It show error in the line for FAISS it says attribute openai has no attributes error
Same D:
@@xsaber69x @virajraina9803 any solution guys?
Can you explain this same example using expressJS? Coz no other tutorial hasn't used expressJS to illustrate this example
What solution can dynamically add or extract database for an LLM?
Like your company information that can be accessible by employees
Hi! I'm trying to use it, but when I write a query in the input box, the chat doesn't show up.
How can I pass in several files at once? I don't just want to use one PDF.
Amazing content. Thank you!
Is there a way to do this with PDFs that have graphics and images?
Hello. Thanks for a great video. But i have financial statement pdf file and it contains tables in it. How can i achieve besy results out of it?
Any suggestions or help would be more helpful. Thanks😊
How long would it take with 50 long 100 page pdf? Which cpu would be needed?
What about using any other open source LLM instead of GPT? thank you!