Видео

thanks! not really on load testing yet, it's quite a niche and advanced topic which mostly depends on the machine where the apps is running, even so, in the next video from that one I showed how to deploy that app on Azure and how to chose the size of cpu and ram of the instance, and how to check the current load on cpu and memory, but I didn't perform load testing.

thanks! 3.10 to 3.12 will work for sure, I'm not sure with 3.13 as it was released 1-2 months ago and some libraries are not yet updated to work with it, I'm still using 3.11 for much of my work and apps.

@@enricd Thanks for the reply. Looking forward for more videos from you.

Barcelona :)

Hi! you can do it with Anthropic otherwise, although you will need to find some other embeddings model. You can also check my older video about building a chat with Google Gemini that also allows to upload docs. I will try to do a video on how to this locally using open source (free) LLMs as well at some point.

And do you know why if we use Ollma, why do we need to use OpenAI? I thought it was enough just with Ollama.

Hey! yes, I will try to do maybe a video doing RAG local with open source models for example whenever I have time for it. Even so, you can check a video in my channel about building a chatbot with Gemini where I already showed how to upload docs into it, which is not exactly the same as the RAG in here, but very similar :)

Thanks! I will try to do at least the AWS video at some point

Hi! What Operating System are you using? Try manually downgrading the onnxruntime python library one or a few versions and checking if this fixes it. For example, if you currently have the version 1.19.2 (you can check your current version with "pip show onnxruntime" or "pip freeze"), then try "pip install --upgrade onnxruntime==1.19.1", and if it still don't works, try 1.18.0 for example

Thanks! I will add it to the list of ideas :) but not sure if I will have time for it any time soon

thanks for the idea! There are some different options out there for this, the easier ones are SAAS startups like Pinecone that hosts the cloud vector store for you, then you can also store it in any cloud storage option and also have a microservice or an api serving it, having the docs there. The thing is that there is not a single solution for this, it depends on your use case, how often the vector store data needs to be updated, how many users you will have, how much docs they need to query, what is your budget, etc.

@@enricd so, you should make video for the rest of this world, Sir! :)

It's possible to use it but it also has some other custom ways of doing it uploading docs to google cloud and also openai and anthropic/claude are preferred for many people. You can check my Google Gemini video in the channel where I showed how to upload docs into it and chat with them :)

You don't want to upload your stuff to the cloud man...not if you value your privacy.....I'm going to use this video but keep it local.

@@AaronBlox-h2t it depends.. I dont mind uploading some cooking recipes pdf to openAI or some public papers to chat with them, if I want a little bit more of security, then I would do what I did in the next video and build this on Azure, and finally if I really want privacy with my sensitive/private/secret data I would do it locally using open LLMs but the result will be slightly worse

😂😂 Thanks!!

Thanks! Try to "pip install cryptography" in the terminal with the venv active if you have created it, and also add it to the requirements.txt file if you want to upload the web app on streamlit cloud or somewhere else. Let me know if that fixes it :)

@@enricd It does, thank you! I'm exploring it and I'm really impressed so far. I will git it a try with 250 1.5 mb pdfs file eventually. Is it possible to configure it so if the requested info is not in the PDFs he simply mention it without using info from outside?

Oups, I get "ValueError: Batch size 664 exceeds maximum batch size 166" when trying to upload 8 PDFS of 1.5mb. Do you know what the limitation are?

About the error, it seems to be in the OpenAI Embeddings model, you would need either to upload them in small groups, or to change the loading function so it doesn't try to load them all at once to the vector store, but one by one. Not completely sure about it, but it's my guess.

I'm not really sure if I get the question, do you want it to answer only using info from the RAG PDFs (not it's own learnt knowledge), or do you mean to go to the internet or ask you whenever the info it needs for answering your question is not in the RAG PDFs?

do you mean my computer parts? for these demos you don't need much, the LLMs run on the cloud from the OpenAI API and the RAG with few files doesn't take much RAM or CPU, I have a pretty powerful desktop PC with Windows 11, but this can run in almost any device with Python 3.11 or similar. If you are in Mac or Linux, maybe few commands or Python libraries may be slightly different, but 99% of it would be the same

not sure, in few days I think I will upload a video doing something similar but in another game/app, and maybe in some weeks/months I will check the rules and regulations in the more popular poker online platforms about using bots and assistants, and if I see it clear I will try, but I think they don't like you to do this kind of stuff :/ and I also never play poker :)

Managed to get it working in the end and have even made a basic interface so you can select the table and turn indicator area. Currently experimenting with pre programmed pre-flop decisions to save on API costs. Been really enjoying this project so thanks for the inspiration.

@@superiorchaos awesome! thanks for sharing, there is so much room for improvement and building features on top of it. In addition, I'm pretty sure that LLMs aren't the most optimal solution or algo to play poker, it would be great to mix the more optimal strategies with LLMs somehow to extract the best of both worlds.

thanks @sitheekmohamedarsath ! it's in the oven :P

Thanks! This week I plan to upload a video on how to do RAG with LangChain and GPT-4o and Claude 3.5, with docs like PDFs, .docx, .txt, markdown and websites content

@@enricd Sounds amazing, I really prefer your method of talking through and just about the right pace for someone like me ith limited coding knowledge. For this model could you just add PDFs to the gemini LLM or would you need langchain or llma index to incorportae PDF access for this omnichat?

Kindly Use Google Gemini Also if possible it'll be helpful ​@@enricd

muchas gracias Leonidas! escríbeme a contact.enricd@gmail.com y le echamos un ojo :)

Hi! I don't know much about poker beyond the basics, but you can probably tune this behavior in the prompt instructions, telling the bot for example to be more conservative, more risky or some more elaborated instructions to perform in a certain style as you desire. At first maybe won't work perfectly, but then you can iterate that prompt to improve it little by little until it works as you want. It would be nice if you try it and comment here how it goes :)

@@enricd it would definitely help out as a tool to give you the best technical analysis of each decision. I don't know if I would ever be comfortable letting it make the choices for you. Mainly in a cash game where everything you bet is your actual money. But there are many tournaments where you just buy-in for the cost of the entry fee which sometimes are for 1 dollar or less. With that style I would be willing to see how it can perform all on its own. Plus this video is 9 months old. The poker skills have likely greatly improved since you made this video. Unfortunately I'm not anywhere close to the skill level needed to program this on my own. Do you have a copy of the code to just use it based on manual inputs and skipping the screen capture parts? When I have asked ChatGPT about NLTH before it didn't seem that skilled in it enough to be able to play autonomously. Is there a certain prompt or version required in order for ChatGPT to provide the best decision making the most up to date model that works?

Hi aayushsmarten, here you can fin the OpenAI API docs: platform.openai.com/docs/guides/vision . As far as I know, there are only these 2 options to send images to the API, the direct URL and the base64 encoding of it. :)

done :) ruclips.net/video/abMwFViFFhI/видео.html

done :) ruclips.net/video/abMwFViFFhI/видео.html

totally.. :/ even so you can try to refine the chain of getting one partial output into the next input with proper instructions and concatenating everything in the end, not easy to do but with enough rails and treaks could works I would say :)

@@enricd hmmm maybe... Would take much API coat just trying to get a good result haha and I think it's not worth it for me at the moment.

I mean Multimodal queries

As a reference, an image of 1000x1000px would take 1400 tokens, so 1000 image would cost 4 dollars. You can get more info in here: docs.anthropic.com/en/docs/build-with-claude/vision

the model itself from the api I would say it's not capable of storing files or keeping them, you have to manage such functionalities in a backend logic in any cloud or your own computer. Like doing a vector database for example and the RAG logic.

no code for download ! :) sorry

Thanks! when you deploy your app into the Streamlit Community Cloud, the <your-sub-domain>.streamlit.app URL is available from anywhere in the internet, I have some other videos in my channel showing to do it. If you want to deploy it somewhere else you can do it either from your own server (which could be risky as you are opening it to the public internet) or you can also deploy it on AWS, Azure, GCP, Heroku and some other public clouds.

@@enricd thanks. for anthropic I get an error about billing "credit balance too low.." . I did create an account and succesfully claimed the $5. Is this not sufficient?

Thanks! GPT-4o doesn't produce images, it could generate the prompt for dalle-3 or any txt2img model to generate them. If you want to send it a document, you could use the langchain document loaders for it :)

totally! that's something I had in mind, but it would add some more complexity (although the interface and integration would be better) and also when dealing with Gemini 1.5 2 weeks ago, I found that langchain was not yet capable of dealing with upload files like videos, audios, docs and so, so it was still for the Gemini 1.0 capabilities but not for the 1.5. But yes, it would have some advantages and depending on the project, I would use it and also structure better the code with more abstractions and a better architecture. Here I wanted to show a basic and plain integration of it.

Hi Vincent, I think whenever you create an account with an email, you already have the $5, but I'm not sure. Try with a new email, and check if this is also available in your country as maybe it's only available in certain regions. I hope you will be able to fix it :)

i have the same problem. i verified my phone number and still couldn't claim it

@@maiseja9987 support.anthropic.com/en/articles/8994925-how-do-i-claim-my-free-credits this is what they say in their FAQs

@@enricdthat didn’t work for me and their support is unresponsive. My phone number is now registered, I cannot re register it, and no credit.