Hi, pretty cool stuff and very helpful! You mentioned the rag approach. A follow up with this approach would be very cool😊. Thanks a lot for your work!
Hi @TimHoffeller, yes RAG with LangChain would provide a more scalable solution. Luckily, my next video will be covering that very topic with LangChain, Supabase and a local Ollama. Check back in the next few days and let me know your thoughts. Thank you for the comment.
Thanks, here are the links, also available in the video description now. github.com/pillarstudio/standards/blob/master/reactjs-guidelines.md gist.github.com/nlaffey/99fdb37c0ba286f38a0582564061dea8
one thing in case of bigger model u need it for more memory in tha case vram is an issue and u need to quantization but then accuracy defects using 8bit or 4 bit also in that case code errors happen , smaller cap are usless due to being applicable in real life projects , but good
Yes, for bigger models, VRAM and GPU can be highly beneficial. And you're right, bigger models will deliver better results for real life projects. Thank you for the comment.
Yeah, that would be cool. Based on the user prompt, the meta agent creates its own tools and executes them as needed. Controlling the agent will be a beast
@@genzprogrammer Easy. What would the RAG do - are you thinking a chatbot to talk to the codebase or something more advanced, like artifacts generation?
@@AISoftwareDeveloper Thinking of build something like a Chat where it takes my query and search the vectordb for which file should i change Nd what should i change
@@genzprogrammer that's a great idea. So, it will tell you what files need to be updated based on a feature change you're thinking. Can you give me one or two examples of queries you'd ask?
@@AISoftwareDeveloper Some tuning would be great, cause the quality of code (that React components) is so bad tbh. I asked several frontend engineers to look at it and it was not the best one. Maybe it's cause of models I used (llama 3.2-3b and 3.1-7b). Also, that would be great to know how to prepare a proper guideline for a model to use, cause links (event to GitHub raw MD file) are not working when there are a bunch of other links on that page.
Awesome video. Can you teach me how to convert this coding to .py file so I can run it in my local machine visual studio code app and check with other compatible models
Thank you for the comment. Yes, you can save a Jupyter notebook and run it as a Python scrip locally. Watch this to learn how: Jupyter Notebooks in VS Code on MacOS ruclips.net/video/3pbFb7X2ObU/видео.html
@@AISoftwareDeveloper I am using windows machine. Hope that video does covers it as well. And thank you so much for a quick response. Really appreciated
You'd want to append the tool response into the messages array. You can then filter that array by instance type, passing ToolMessage as the filter. That will do the trick 👍
Hey, definitely. You can use this and create your own Github Copilot as a VSCode extension. Here's how to get started: code.visualstudio.com/api/get-started/your-first-extension
Here is the source code repo: github.com/aidev9/tuts/tree/main/langchain-ollama
Hi, pretty cool stuff and very helpful! You mentioned the rag approach. A follow up with this approach would be very cool😊. Thanks a lot for your work!
Hi @TimHoffeller, yes RAG with LangChain would provide a more scalable solution. Luckily, my next video will be covering that very topic with LangChain, Supabase and a local Ollama. Check back in the next few days and let me know your thoughts. Thank you for the comment.
@@TimHoffeller the RAG video is now available. Any feedback is appreciated.
Great explanation and easy to understand example. Thanks for sharing your knowledge.
Thank you @arunbhati101, I am glad you got something out of it. What videos would you like to see in the future?
Cool video. Source code is great . Also including the relevant urls in the video would be good. Thanks.
Thanks, here are the links, also available in the video description now.
github.com/pillarstudio/standards/blob/master/reactjs-guidelines.md
gist.github.com/nlaffey/99fdb37c0ba286f38a0582564061dea8
one thing in case of bigger model u need it for more memory in tha case vram is an issue and u need to quantization but then accuracy defects using 8bit or 4 bit also in that case code errors happen , smaller cap are usless due to being applicable in real life projects , but good
Yes, for bigger models, VRAM and GPU can be highly beneficial. And you're right, bigger models will deliver better results for real life projects. Thank you for the comment.
what it would be nice, is to have a meta agent that creates dynamic tools and reinserts them into the flow when needed
Yeah, that would be cool. Based on the user prompt, the meta agent creates its own tools and executes them as needed. Controlling the agent will be a beast
ollama models by default are configured to have 2k context size, fyi
Thanks, that’s great to know.
Thanks Bro, 🎉 Correct Time. Was looking for something like this
Thanks for the comment. What would you like to see next?
@@AISoftwareDeveloper How can we load a complete codebase from a git repo and Implement a Rag for that codebase?
@@genzprogrammer Easy. What would the RAG do - are you thinking a chatbot to talk to the codebase or something more advanced, like artifacts generation?
@@AISoftwareDeveloper Thinking of build something like a Chat where it takes my query and search the vectordb for which file should i change Nd what should i change
@@genzprogrammer that's a great idea. So, it will tell you what files need to be updated based on a feature change you're thinking. Can you give me one or two examples of queries you'd ask?
Hey, this was surprisingly helpful!! Thank you so much!
Can I ask for a source code please?
Here you go: github.com/aidev9/tuts/tree/main/langchain-ollama
Thanks for the comment. What topic would you like to see next?
@@AISoftwareDeveloper Some tuning would be great, cause the quality of code (that React components) is so bad tbh. I asked several frontend engineers to look at it and it was not the best one.
Maybe it's cause of models I used (llama 3.2-3b and 3.1-7b).
Also, that would be great to know how to prepare a proper guideline for a model to use, cause links (event to GitHub raw MD file) are not working when there are a bunch of other links on that page.
Here is the repo: github.com/aidev9/tuts/tree/main/langchain-ollama
Awesome video. Can you teach me how to convert this coding to .py file so I can run it in my local machine visual studio code app and check with other compatible models
Thank you for the comment. Yes, you can save a Jupyter notebook and run it as a Python scrip locally. Watch this to learn how: Jupyter Notebooks in VS Code on MacOS
ruclips.net/video/3pbFb7X2ObU/видео.html
@@AISoftwareDeveloper I am using windows machine. Hope that video does covers it as well. And thank you so much for a quick response. Really appreciated
The video doesn’t cover windows, but once you have VSCode installed, running Python is the same. I hope that helps. Thank you for your comment.
Nice man 👍🏻, just wanted to where should i append my tool/function response, like in system or in tools😅
You'd want to append the tool response into the messages array. You can then filter that array by instance type, passing ToolMessage as the filter. That will do the trick 👍
Next step? using voice to give it the prompt :-)
Nice one, @agraciag
Build it in 10 minutes, waste days in the future dealing with dependency issues
Not to worry, mate. We’ll have AI to fix all dependency issues 😉
Sounds like someone has SDE …. Small developer energy😂😂🫰
@codelinx
XD
I hv one question can i somehow create github co pilot who just giveme suggestions wht i need to do using this ?
Hey, definitely. You can use this and create your own Github Copilot as a VSCode extension. Here's how to get started: code.visualstudio.com/api/get-started/your-first-extension
@@AISoftwareDeveloper thanks sir I will try this will it give me suggestions like an co pilot give can I publish that extension as well