instead of putting the files in the project directory can we upload the files in AWS S3 or other cloud storage? how can we handle the conversion of files to vector embeddings using that approach?
I don't understand why to use langchain! It make simple things became complex! And also remove all freedom the programs have because of this "chains"! But, in other hands the TOOLS are the best ones!! I just use the tools! forget about this complicated chains and rules to use things and "this must do in this way", that's not why programming was made for!
Thank you for the video, it was incredibly helpful. I was wondering if you could create a video that demonstrates the integration of all these elements. It would be fantastic to see how to set up a conversation, that will be stored, that utilizes sequential chains, incorporates the Vector Database, and implements an Agent. Watching all these components work together would be truly insightful. Managing chains, especially in larger projects, is a challenge I'm trying to overcome.
Can I only use FaissStore with the paid version of the API key? I am getting an error: Error processing documents: Error: Request failed with status code 429
You are amazing. I'm a web dev, and I found your channel because of the first langchain tutorial. Then I find this, you have thought of everything! Truly grateful, and I'm keeping a close watch to learn from you. For your info, I would most certainly pay for a course by you on bringing an LLM app to production.
@@codingcrashcourses8533 Thank you, yes, I've had time to go through the rest of the videos. Truly excellent. I have a lot of learning to do to get up to the right level, and will keep coming back for more.
Is it possible for you to make a tutorial on how to use langchain JS with Amazon Bedrock API? I don't see many tutorials on the same. If you could do that, it would be super helpful. Great video by the way!
Nice and clear video. Maybee something for the future: I'd like to know more about how to create a custom tool, like creating a database connection and sending SQL as input for example. I tried using the calculato.jcs file as a template for a new custom tool but i get all kinds of errors from the library i dont understand unfortunately
Well this is an advanced request not many might be interested in. I also have other, more project based videos on the channel. This is just an Intro to the lib
excellent course, covers the basics to get started, i prefer the frame work in Javascript Node.js although i know the libraries are the same as the python one
REST Services should be stateless, so you should handle the memory outside of that service. I recommend not using the memory class inside a REST Service. I created multiple videos on langchain in production and one makes use of redis as memory. Here is the video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-I_4jEnDwGwI.html
Do you recommend any free to use LLM? OpenAI is no longer free and I just would like to play around langchain. Also, how safe are the LLMs to use in terms of getting our data exposed for their own ML scripts? I mean, will ChatGPT be trained by my chain calls using my private database resources?
Llama2 is currently state of the Art for open Source models. For openai models you can opt out, but i would recommend against using it for sensitive data.
@@codingcrashcourses8533 wow, really appreciated the llama 2 recommendation. I'm reading some articles and it looks amazing. I'm just wondering about the model being trained using our private data as well. I'll check if I can find an optout as you mentioned OpenAI does. Thanks again!
Hi. Can you please help select from LangChain objects? I need functionality from LangChain that allows me to switch a prompt (or probably a chain) if the user asks "How do I pay less tax?". If he asks this question, I need to ask him about his form of ownership, type of taxation, etc. and then use this information to select another prompt for further processing. But I use ConversationalRetrievalQAChain to work with OpenAI and Pinecone. My question is what is better to use for such a task? SequentialChain or Router or something else? Please not that I use TypeScript to write this app. Thank you!
I have the same task (with python) and do it like this. I use a custom classfier chain that returns an object. I run normal LLMChains and add the documents from the vectorstore manually, since it gives me more control, but thats of course up to you :)
@@codingcrashcourses8533 Sounds not so easy for me, especially considering JS/TS langchain isn't the same as Python version (not all of the objects are there. I couldn't find Router for example). Could you maybe shot a tutorial about this Dynamic Prompting with your custom Chain class or with some langchain object? :) I can't understand how those other chains would have the same "memory" as my default. Especially if simple chains doesn't have Vector store parameters in its initialisers.
@@wordpressmagics2613 Yeah, Router Chain seems not to exist (yet). Well, all these chains in the end help you to create a final prompt for the LLM which you send to the API. If you understand what the API expects and what the Interfaces of the Classes do under the hood, you can choose if you want a more high level chain or not. In the end, you can also build the history yourself if you want
@@codingcrashcourses8533 I have installed langchain using: npm install -S langchain However, when I import I get the following error. I have put the target in tsconfig as ES2020, too: ../../../../node_modules/langchain/dist/callbacks/handlers/tracer.d.ts:22:23 - error TS2583: Cannot find name 'Map'. Do you need to change your target library? Try changing the 'lib' compiler option to 'es2015' or later. 22 protected runMap: Map; ~~~ Found 1 error in ../../../../node_modules/langchain/dist/callbacks/handlers/tracer.d.ts:22
Late for the party, but the error is in your import. I suppose you're trying to "import {callbacks} from 'langchain'. But thats deprecated. You should do something like 'import {handlers} from 'langchain/callbacks' This is not a copy and paste solution, but just to express the idea that now you have to import from the individual modules instead of root 'langchain'