That’s great. Currently, I watch all videos from the series and I have lots of hope to see particular things. 👍 Every video leads to more questions. 😄
@leonvanzyl
4 ай бұрын
😂
@dervla0412
Жыл бұрын
Keep these coming please. EXACTLY what I’ve been looking for. Looking forward to deployment vid. Thank you!!
@leonvanzyl
Жыл бұрын
Thank you for the donation 🙏. Just glad I could assist you in some way 😊.
@emmablyth3580
3 ай бұрын
Such a clear, brilliant series. Exactly what I needed. Thanks so much!!
@leonvanzyl
3 ай бұрын
Thank you
@PepettosClub
9 ай бұрын
Hello Leon, I really appreciate your videos on Flowise, which explain the concepts in a very clear and accessible way. It would also be interesting to see how wizards can be set up using Chroma, as well as an open source Ingest system.
@altonjones5271
Жыл бұрын
Wow - what a great tutorial on Flowise - all clearly explained and easy to follow. Great job Leon!
@leonvanzyl
Жыл бұрын
Thank you 😊.
@Ericperezg
Жыл бұрын
Thank you very much for the videos! I have discovered Flowise and put it into practice thanks to your videos. Above all, very well explained and detailed. It would be interesting to have a second part explaining how to maintain these documents. Being able to delete documents that I want to have in the database. And thus maintain an updated database.
@kevinseale4080
Жыл бұрын
I second this request and also say thank you for the very helpful videos.
@jayabalanbabu1613
8 ай бұрын
I echo, @Ericperezg. I am also look for this. For example, suppose I have two web pages, A and B, embedded in Pinecone. If I make changes to page A, how can I update its corresponding embedding in the Pinecone index?
@mikecorbean9985
Жыл бұрын
phenomenal LvZ! Your work is not going unappreciated my friend.
@leonvanzyl
Жыл бұрын
This means a lot 😁. Most of the time I think that I'm the only person geeking out over these tools. Just glad I could help ☺️
@mikecorbean9985
Жыл бұрын
@@leonvanzyl we're right here w/ you 😉
@RobertoFabrizi
9 ай бұрын
About the production ready document chatbot, there is a new injestion button now that is specific for that task, and simply asking questions to the CRQA chain doesn't trigger a new injestion every time.
@adriaanb7371
Жыл бұрын
off the scale quality videos! Just need to wrap my head around my newly acquired super powers 😉
@leonvanzyl
Жыл бұрын
Glad I could help! With great power..
@PeterPanbe
Жыл бұрын
Looking forward to the rest of the series, including how to actually put this in the cloud and build a public chatbot
@leonvanzyl
Жыл бұрын
Next video will be on deployment. We'll definitely also have a look at calling the Flowise APIs from our own applications 👍. Just glad you like the videos man.
@DeepakVijaykumar-v9i
Жыл бұрын
@@leonvanzyl Hi, Thanks for the excellent video. Can you please advise how to use the Folder Path in "Folder with Files" document loader. I've deployed the Flowise app on GCP as docker image. This is a deal breaker as I have 100's of pdf documents that I want to feed in to the flow. Thanks
@Christopher_Tron
Жыл бұрын
Thank you for this content! I truly appreciate you helping me learn this great tool. Also your teaching style is great, looking forward to the rest of the playlist!
@leonvanzyl
Жыл бұрын
Thank you very much for the support and compliments. I've been recovering from surgery, hence the lack of videos. However, we're back on schedule and the latest video will release in a few days.
@abiolasamuel8092
7 ай бұрын
You're doing an amazing job. Can't wait for the updated video! 👍
@leonvanzyl
7 ай бұрын
This is incredible, and your support keeps the channel alive. Thank you VERY MUCH 🙏
@charasmaticcrystal
Жыл бұрын
Thanks for creating the easiest flowise videos.
@leonvanzyl
Жыл бұрын
Thank you
@Alex29196
2 ай бұрын
Thanks Leon. love your tutorials.
@armanabdullah1670
Жыл бұрын
Your tutorials are amazingly helpful ❤❤❤🔥❤🔥
@leonvanzyl
Жыл бұрын
Glad I could help 😃
@JonathanLyon
Ай бұрын
Hey Leon, love your work! Is it possible to use llama-parse in flowise for complex and hefty PDF's? If so, any chance you could do a video on the subject please? Thanks Jonathan
@BuddhikaNelum-zx9sl
4 ай бұрын
Thank you for creating this great tutorial! I have a question regarding database connectivity. I'm attempting to connect my chatbot to a local MSSQL database using "SQL Database Chain" but I'm struggling to establish a successful connection. Despite trying various connection string formats, I haven't had any luck. Could you guide me on how to correctly format the connection string for this purpose?
@AssassinUK
Жыл бұрын
GREAT video. Next time consider including META data in the Upsert. Subsribed!
@leonvanzyl
Жыл бұрын
Thank you!
@cargobridgesolution978
Жыл бұрын
Thank you so much for your tutorials, I'm following each of them. I have currently difficulties to connect stock website API... Hope I can find a concerned tutorial from your channel😊
@leonvanzyl
Жыл бұрын
Thank you for the feedback. What is the issue that you are experiencing?
@TheDudeWithSome409
Жыл бұрын
I tried experimenting with different vector stores and found that the different databases all behave the same. However, when I changed the recursive character text splitter to a chunk size ot 2000 I received much better answers from documents when I ask specific questions on technical information. I do not understand why 200 in 2000 created such a difference in its capability to respond to direct questions from the text itself.
@leonvanzyl
Жыл бұрын
I've noticed that as well. It all depends on the context I guess. The example I used uses very short and factual sentences, so the model was able to easily find the answers within a 200 character limit. For more advanced examples, where you might need an entire paragraph to make sense of the data, 2000 characters might be required. You're not doing anything wrong. This is all about playing with that number until the results make sense.
@ThorOnYouTube
8 ай бұрын
Thanks for pointing this out. I was trying to summarize a news article and was getting an inaccurate result even though I set the ChatOpenAI temperature to 0. When I explored the vectors on Pinecone's GUI, I found that one of them held a text snippet that was quite short and vague (without its nearby words in the article). I deleted the vector and, as expected, the summary improved, but was still a bit too high level. I found that by increasing the splitter's chunk size to from 200 to 500 and overlap from 20 to 50, I got an excellent summary. (First, I deleted all vectors and then upserted again.) Hi youtube.com/@leonvanzyl, any guidance on how to do splitting between semantic clusters so that text doesn't get fragmented into meaningless phrases? Or should a larger overlap help with that?
@TheDudeWithSome409
Жыл бұрын
Excellent video You have the best tutorials out there on FloWise! It would seem that you have a niche right now for KZitem videos after seeing a number of other people in this niche if you continued to make videos you would have over 50K subs in a short time you would find that you would hit a plateau in a wall around 75k. Not saying you couldn't break through that wall but I haven't really seen this space of LLM breakthrough very often. I was wondering what is your long-term goal for the channel?
@leonvanzyl
Жыл бұрын
Thanks for the compliment and feedback 😁. Generally speaking, this is a coding tutorial channel with AI in the mix. I'm definitely not limited to LLM videos, but I do see that AI-related coding courses are here to stay. This comment was food for thought though. With that said, I would be incredibly humbled to reach 50k subs.
@TheDudeWithSome409
Жыл бұрын
@leonvanzy Your videos are very well edited, which makes them a cut above the rest. I was looking for resources to build out internships at the charity I work at, Deocracy Institute, and Hopefully Flowise will be the right tool to allow for people to add to some of the projects we are working on. We have been sharing your videos around as they are so helpful and informative. One of the projects I'm hoping to be able to integrate Flowise with is building out tools to help a reporter have their content rewritten for different audiences, build reports by scraping the internet, and other helpful features a reporter would use.
@leonvanzyl
Жыл бұрын
@@TheDudeWithSome409 Thank you for the compliments and for sharing my videos. You have sincerely made my day!! It's awesome that you're doing charity work. Respect my friend.
@rorychockman1052
Жыл бұрын
Awesome video! I have a question? If I upload a document which itself is updated every week, will the data in the vector store also be updated? or will it simpley add to the data that is already inside?
@leonvanzyl
Жыл бұрын
If you upload the exact same document again, it will simply add the new entries to the database. This means you could have duplicates. There is no real way for the vector store to update existing entries like a traditional database. You'll have to get creative, but the simplest solution is to simply clear the index of the old entries before uploading the updated file.
@BestRestaurantToEat
10 ай бұрын
@@leonvanzyl How do we clear the index, and also instead of using pinecone as a database can we use traditional database like MySQL to store the vector information
@BJM1896
Ай бұрын
Leon, please show a document retrieval directly from Pinecone to an agent in a Multi-Agent flow. Not to a chain, to an agent. Not Sequential Agent, Multi-Agent, please.
@drtunmyatwin
5 ай бұрын
Pinecone authorization error API keys are working and checked again and again what is the problem. Could you please help me?
@calixtohyj
9 ай бұрын
Thank you, I love your videos
@leonvanzyl
9 ай бұрын
Thank you 😊
@sonalisen7912
9 ай бұрын
Hello, your videos are great. Can you share any step by step tutorial on flowise with chromadb?
@jonm691
7 ай бұрын
Great video - I note that now (Jan '24), the upsert functions will be deprecated. Will this change the need for separate ingest and query parts?
@leonvanzyl
7 ай бұрын
I would suggest separating them in real world scenarios as well 👍.
@jonm691
7 ай бұрын
Thanks - good advice. @@leonvanzyl
@nelsonm6075
2 ай бұрын
Nice. But, in upsert, file docx, error "end of data reache (data lenght=0, asked index=4) Corrupt zip? Strange, not a zip file, an docx file. To try pdf file, error "pdfdocument : strean must have data". Detail, my documents have some images. Would I have to remove all the images for it to work?
@mikecorbean9985
Жыл бұрын
Thanks!
@leonvanzyl
Жыл бұрын
Wow, thank you for the super! Much appreciated 😊
@peterpaul6953
3 ай бұрын
Thanks for the video! But how would you go on with more than 1 document and even different sources? Connect them all to the same Pincone or create a new node for every document? and how to combine them all in the end?
@leonvanzyl
3 ай бұрын
You can add multiple document loaders to the same canvas and vector node.
@ai-but-autopilot
Жыл бұрын
AWESOME TUTORIAL
@leonvanzyl
Жыл бұрын
Awesome comment!
@bourbe
11 ай бұрын
Hello Leon, Fantastic value sharing !! Thanks a lot. I have a question please : How to adapt the first flow for a csv file ? I will use Csv file insteas od Text File but WHAT I am supposed to use for "Recursive Character Splitter" in the case of a csv file pls ?
@leonvanzyl
11 ай бұрын
Thanks Clark! I think there might be a CSV splitter actually.. will check it out
@bourbe
11 ай бұрын
@@leonvanzyl OK thanks in advance' I will check too
@senorchavo5572
6 ай бұрын
Awesome tutorial, thank you for the information. there isn't update now for flow wise, and for some reason, I cannot Get pine cone to work properly... Ends the chatbot cannot read the document Is there limitations with the free version of pine Cone? Is there a certain way I need to configure it?
@leonvanzyl
6 ай бұрын
Hey there. I have created an updated Flowise series which you can find on my channel 👍
@eumarcoaureliorosa
Жыл бұрын
First congratulations for the work, this series is very good. I would like to know when do you think you would need a Pinecone paid plan?
@leonvanzyl
Жыл бұрын
Thank you for the feedback 🙏. The free database gets deleted after a week of inactivity, so you definitely don't want to use that in Production.
@nabiulfarhannabil5646
Жыл бұрын
Hey. thanks for the detailed video. One ques, how can I use chromadb? as chroma doesn't have any API or credentials.
@leonvanzyl
Жыл бұрын
You're welcome 🤗. You can simply add the ChromaDB node to the canvas and use it like the Pinecone node. ChromaDB runs locally and does not need any credentials.
@waynenunn7319
Ай бұрын
Leon, please update this tutorial with the complete setup of Pinecone. Thanks, Wayne
@leonvanzyl
Ай бұрын
Hey Wayne, check out the 2024 playlist instead 😊
@waynenunn7319
Ай бұрын
@@leonvanzyl Thanks, I missed that one.
@eliasnordenberg4449
Жыл бұрын
Does this work with the free pinecone version? Great video btw:)
@mikecorbean9985
Жыл бұрын
Yes it does but as Leon noted previously, "The free database gets deleted after a week of inactivity, so you definitely don't want to use that in Production."
@rosenareku5071
Жыл бұрын
Great tut! How can we link our domain in the URL. It would be great to learn how to change from localhost to the domain link for a locally hosted server having a public IP address. Otherwise, you've made my day as l perused through all your Flowise AI videos.
@leonvanzyl
Жыл бұрын
We'll definitely cover deployment next 👍
@nathanbui919
4 ай бұрын
How would you implement this chatbot to your website to handle file uploads from the user?
@LesHenderson
Жыл бұрын
I would also like to know how to add multiple documents via upsert. Is it one at a time? How do you keep track of what you uploaded? Can you delete data from older records at some point?
@leonvanzyl
Жыл бұрын
You can definitely upload more than one file. You can attach multiple document uploaders at the same time, or even use the "folder" document uploader. There is no clear answer on how to keep the knowledge base up to date. There is no way to "update" the existing records, and there is no way to clear the index from within Flowise either. I've been trying out different techniques and hope you create a video soon. For now, just clear the index manually and re-upload.
@WayneBruton
8 ай бұрын
Hi Leon, hope you are well. Great video, quick question, I uploaded my cv as a test. When I ask it what my key skills are, it answers in Spanish,
@leonvanzyl
8 ай бұрын
You can prime the modem by clicking on Additional Parameters on the agent. In the system message, add something like "respond in English"
@WayneBruton
8 ай бұрын
@@leonvanzyl Thanks Leon. Quick question, I have just gone through your tutorial where you are using replicate. Does llama have a model where you can upload a document and query it? A bit like open assistants?
@SimmentalerCow
9 ай бұрын
How do I use a CSV instead of a text file or a docx file? Do I need to change the text splitter? Thanks.
@adhi8005
Жыл бұрын
Amazing Video Series. Thank you. Just curious, how do we store the vectors in our own database apart from the ones available in vectorstore?
@leonvanzyl
Жыл бұрын
Thank you! These vector databases are quite unique and a "normal" database will not work.
@jacobriedel5326
Жыл бұрын
So the first time you run it, it will upload the vectors, do you create a separate flowise app to do these jobs (one time upload jobs) so that your Chatbot only has to read the pincone index? I am just curious if there is a way to do some of these tasks all in one app instead of creating multiple per use case.
@leonvanzyl
Жыл бұрын
The example in the marketplace will upload and read in the same chatbot (I think that is what I did in the previous video in the series as well). This does have its disadvantages, but for small, ad-hoc use cases it should be fine.
@sathyanarayananramamurthy5681
3 ай бұрын
can we upload a PDF directly to the redis vector ?
@KotokiTokito
Жыл бұрын
Hello, thank you again, your rythm is so good! I am wondering if it is possible to finally have a local application to chat with the doc stored on pinecone.
@leonvanzyl
Жыл бұрын
Thank you for the compliment!
@mello1016
Ай бұрын
How can we store the agents output to a vector database not just retrieve it?
@BJM1896
Ай бұрын
Thanks!
@leonvanzyl
Ай бұрын
Thank you for the super!
@MichaelAverto
Жыл бұрын
Great tutorial! Is there a way to take the user's input from the chat and provide it back into the chain for learning? I see the conversational retrieval QA Chain has an output but not sure if this is supported yet?
@leonvanzyl
Жыл бұрын
Thank you for the compliment. Not 100% sure I understand the question, but you might want to add buffer memory to the chain. That way the conversation history will be taken into account in the chat.
@josephn7392
Жыл бұрын
im wondering how i could prime the bot to respond to a particular data set with a specific tone. This would be helpful for making a chatbot that can query data and provide feedback t the user in a more natural language form.
@leonvanzyl
Жыл бұрын
I'm going to cover prompt chaining in this series. Effectively, we'll pass the response of our document chatbot to another chain that can format the response in a different style, tone, etc.
@thorbjrnstrmberg
Жыл бұрын
Greate content! I really learn much from your videos! I get a request failed code 400, do you know how to fix it?
@TuxDoido
Жыл бұрын
Congratulations for the class, your content helped me a lot. If you can, I would like some help on how to store chatgpt in the database in this case, also remembering the context of the conversation. Would that be possible?
@leonvanzyl
Жыл бұрын
That is actually a great idea for a larger project whereby we store the conversations in a database. Thanks.
@TuxDoido
Жыл бұрын
@@leonvanzyl thank you so much for your videos, you helped me a lot!!!! Gratitude
@thegptfamily
Жыл бұрын
Thank you so much for your video. Question: is there a difference if .txt files are easier for the model to learn versus pdf? Having difficulty with the model providing a lot of answers of it not locating the answer according to the context.
@leonvanzyl
Жыл бұрын
You're welcome. Text files are definitely easier to process than PDF files.
@flavioguzman
Жыл бұрын
This is amazing.
@leonvanzyl
Жыл бұрын
Thank you!
@View6000
11 ай бұрын
I'm getting the error 'this.retriever.getRelevantDocuments is not a function'. Any solutions?
@famfloro
11 ай бұрын
You generated a summary of the story by asking for it, right? How is that the model can generate a summary out of the top 4 matching document chunks? And, the story doesn't mention anything about summaries. How is that the system selects the top 4 most relevant document chunks?
@abiolasamuel8092
7 ай бұрын
Thanks for these Flowise video series. So I followed the steps outlined, uploaded a pdf file and used Pinecone starter plan (the plan used in the video is not free). But when I ask the bot a question based on the pdf file, it couldn't answer. When I asked it to summarise the pdf, it says "Hmm, I'm sorry, but I am unable to access or view any attached files as I am an AI text-based assistant". Any solution to this?
@leonvanzyl
7 ай бұрын
Hi there. The RAG solution has changed a lot since I recorded this video. Will publish an update to that video in the next few days 👍
@ProfessorSaibertin
7 ай бұрын
Thank you for the tutorial. Question: if I want my data source to be updated every day via another API (so the API updates the source file/source data), how can I do that?
@leonvanzyl
7 ай бұрын
You could schedule a Cron job that runs every day and triggers that API. Cron-job.org is free to use.
@ProfessorSaibertin
7 ай бұрын
@@leonvanzyl thank you for the quick info. What do I search on KZitem to learn how to do this exactly? (no-coder here)
@bshhun
Жыл бұрын
Hi! Great video! I have a setup similar to yours. How can I prevent the bot from answering off-topic questions? I've loaded veterinary data into Pinecone and don't want the bot to answer questions on other topics. I added this to the System Message in the Conversational Retrieval QA Chain, but it had no effect.
@leonvanzyl
Жыл бұрын
Thank you for the feedback. That is really interesting... the agent should only answer questions based on the knowledge in the Vector Store. When you click on Additional Parameters on the agent you can change the system message. You could try instructing the AI not to answer questions that I are found in the knowledge base.
@simon.stjohn
Жыл бұрын
You mention in this example we could use the OpenAI Chat or Open AI LLM and its up to us to choose but I am not quite clear the difference. I know the Chat model is designed for chat, but in what cases would we choose the OpenAI LLM?
@leonvanzyl
Жыл бұрын
You get access to different models. These models have different pricing and performance. For instance, if you simply want to classify data, there is no need to use a complex model like GPT3 or GPT4.
@nntun03
Жыл бұрын
Thank you so much for the detailed explainations, your videos are very helpful!! How can I add a calculator pls? the llm goes awry when i ask 'how many months did he work there', even tho I provided all the months of work..
@leonvanzyl
Жыл бұрын
You're welcome 🤗
@sandowssecret7363
Жыл бұрын
Great video thanks so much Just wondering how could store and retreve chatbot conversations?
@leonvanzyl
Жыл бұрын
You're welcome 🤗. Can you clarify? Are you referring to something like writing the conversation to a file?
@sandowssecret7363
Жыл бұрын
@@leonvanzyl yes or even have all the conversations from the bot as its own chat so it can be embedded in a website or as a link.
@xavierparadiso
Жыл бұрын
I don't understand how the embeddings node works, it seems like it secretly gets output from the Conversational retrieval node?
@leonvanzyl
Жыл бұрын
Hey there. The embeddings node takes the output from the text file uploader and converts the documents into vectors. These vectors are then stored in the database.
@jonzh4720
Жыл бұрын
@@leonvanzyl have the same question, after embeddings are stored in the database why still need OpenAI Embedding for second jobs? I thought we can read the database directly. This still has to call OpenAI twice one for LLM and Embedding.
@famfloro
11 ай бұрын
@@jonzh4720You want to generate an embedding of the question (a numerical vector) and then find out what is the closest embeddings in the database. You cannot lookup using the raw text question. You need to convert the question into an embedding (vector) first.
@Stinkfingerfilm
Жыл бұрын
Be aware that a free pinecone is storing files for a week. After an week of inactivity it will closes the database. If the use remains it stays active... untill inactivity for a week 😢 then it closes the inxex... I have a question about the video. If i upload several PDFs that are in topic overcrossing each other... and i make more than one chatbot that communicates to the Pinecone. How do i assign each chatbot (q&a) to the right pdf?
@leonvanzyl
Жыл бұрын
Hey there. Yes, the Pinecone DB is removed after a week of inactivity if you're on the free plan. You could look into using Redis or Supabase. For the second part of your message, you could assign metadata to the documents. You can then "tell" the bot to fetch the answers from the documents containing that specific metadata. I'll see if I can create a video on it.
@seakyle8320
Жыл бұрын
How to clear the pincone data? What if i change a document und upload it again? would it replace the old data? How can i specifiy the user imput promt to search only in one of the uploaded documents?
@leonvanzyl
Жыл бұрын
Good question. There seems to be a few approaches to this. The first is to clear the index manually or programmatically before executing the injest flow. Another approach is the assign metadata which included a version number of sorts, and then changing the chatbot to read documents containing that metadata only. This means that the database will grow all the time with data that is redundant / outdated. I would definitely like to hear some other thoughts in the comments.
@seakyle8320
Жыл бұрын
@@leonvanzyl Would be nice to have a video about this topic 💕
@SrikanthNandiraju
6 ай бұрын
How do I use the Flowise Query API to apply a Pinecone filter? I came across a snippet for Supabase but am having trouble converting it to Pinecon e. I would appreciate it if you could guide me on how to modify the Pinecone filter using the Flowise API. This snippet is for supabase { "question": "What's Jane?", "overrideConfig": { "supabaseMetadataFilter": { "supabaseExistingIndex_0": { "user": "jane", "docname": "janedoc" } } } }
@JonathanLyon
Ай бұрын
Thanks
@leonvanzyl
Ай бұрын
Thank you for the support ❤️
@JonathanLyon
Ай бұрын
@@leonvanzyl all good Leon, any thoughts on whether it's possible to use llama-parse in flowise?
@roamingh
Жыл бұрын
How would you change this work flow to have an agent? If the answer is not in the document, then go to the web and search for an answer.
@leonvanzyl
Жыл бұрын
Great idea for a video - combining chains and agents 👍. In the meantime, you can have a look at the "Chain Tool" under the tools section.
@rajneilg
Жыл бұрын
Your series is amazing! I'm trying to follow along but I keep getting the following error "PineconeClient: Error calling upsert: PineconeError: PineconeClient: Error calling upsertRaw: FetchError: The request failed and the interceptors did not return an alternative response". Only difference is that I can only select the Starter option in the Pinecone. Any ideas?
@killer81095
Жыл бұрын
same here man
@killer81095
Жыл бұрын
figured it out...just remove the pine cone ui component -> ctrl+shift+C -> put a new pinecone component seems like the flowise ui is little clumsy ...
@leonvanzyl
Жыл бұрын
Thank you for the compliment! @killer has the right solution. That seems to happen after updating Flowise.
@GREENDAYROKS1
11 ай бұрын
Hey bro could you explain that a bit more, I have the same problem. Do you mean remove the Pinecone Upsert Document? Cheers! @@killer81095
@DeepakVijaykumar-v9i
Жыл бұрын
Hi, Thanks for the excellent video. Can you please advise how to use the Folder Path in "Folder with Files" document loader. I've deployed the Flowise app on GCP as docker image. This is a deal breaker as I have 100's of pdf documents that I want to feed in to the flow. Thanks
@leonvanzyl
Жыл бұрын
Thank you for the feedback! I'll try and cover that loader in a future video 👍
@marceloperazzo2901
Жыл бұрын
Thank you very much for spreading the knowledge. When I do the second chatflow with pinecone existig idex, it doesn't read the context. In other words, you don't know what to say, why would it be? Thank you.
@leonvanzyl
Жыл бұрын
You're welcome 🤗. Very hard to say what the issue could be. Ensure that you enter the correct index details. In Pinecone, ensure that vectors are not 0 (ie. data was uploaded by upsert).
@davidstephen8511
8 ай бұрын
How do we work with this kind of setup in cases where we want to incoporate a system message from a chat prompt along side the document uploaded
@leonvanzyl
8 ай бұрын
You can change the system prompt by clicking on Additional Settings on the agent / chain.
@Yakibackk
Жыл бұрын
very good, why do you not put 1000 the limit 1000 chunk size instead of 200 because at the end of the day its the totalt of the tokens which is matter or do i missing something ?
@leonvanzyl
Жыл бұрын
Keep in mind that these chunks will eventually be included as context in the conversations with the model, and there is a cost associated for the amount of tokens used. You ideally want to keep the chunks as small as possible while still getting good results.
@Yakibackk
Жыл бұрын
@@leonvanzyl thanks
@adityahpatel
Жыл бұрын
Trying to do Q/A with a PDF of a company’s 10-K. I’ve set up a chatflow as shown. Not getting good results..e.g. even for simple questions like what was annual sales in 2022, it is giving wrong answers or saying it doesn’t know when the information is in there…any high-level suggestions?
@leonvanzyl
Жыл бұрын
Hey there. It could be that the PDF contains a structure or font that is not easy to extract the text from. Another issue could be that the document does not contain the answer. Does the document contain the answer for what the sales are, as in, does it contain a piece of text that clearly states "the sales for 2022 is XYZ"? The bot will not be able to perform analytics on the knowledge base.
@adityahpatel
Жыл бұрын
@@leonvanzyl Hi, yes its a compustat equity research report on Apple Inc. It contains balance sheets, income statements, cash flow statements etc. The answers are definitely in there, but its not finding them!
@famfloro
11 ай бұрын
Have you tried to use a PDF text extraction tool to see how feasible it is to answer your questions from the extracted text? Extracting text from a PDF is hard enough on normal files, much more difficult on obfuscated files.
@hackprender1786
Жыл бұрын
I keep getting this error and i cannot figure out why: "PineconeError: Failed getting project name. Error: Unauthorized - API key is missing or invalid for the environment "us-west4-gcp-free". Check that the correct environment is specified."
@leonvanzyl
Жыл бұрын
Guessing there's a typo or something wrong with the Pinecone credentials that you used 🤔.
@Fahad-m2u
7 ай бұрын
Hello Actually I followed exactly same but the AI doesn't know anything about the data. Do you have any idea what is causing this ?
@leonvanzyl
7 ай бұрын
It's hard to say with the information provided. You need to ensure that the data was indeed uploaded to the vector store. Did the amount of vectors in the database increase? You could also try to increase the chunk size in order to improve the responses. Also check for any errors in the console.
@davidmaugrion5442
Жыл бұрын
Could you explain on using Postman with Flowise ?
@leonvanzyl
Жыл бұрын
For sure! Will cover it in the API section of this series 👍
@bhawanaprasain621
8 ай бұрын
Can we upload a folder of around 500 files at a time for vector database?
@leonvanzyl
8 ай бұрын
I never tried that many, but I don't see why not 😁. You MIGHT run into a timeout. Let us know 😁
@bhawanaprasain621
8 ай бұрын
@@leonvanzyl I have lots of documents about sth like office procedures and want to ask questions like which information is on which file, how to carry out a process x in office and so on.
@16bitjack16
Жыл бұрын
hmm... I run the bot but it gives me a "blob is not defined". Did anyone else have this issue?
@leonvanzyl
Жыл бұрын
Be sure to save the flow before running it. I've seen that message when I forgot to save.
Пікірлер: 168