Easy RAG Setup - Load Anything into Context - Mistral 7B / ChromaDB / LangChain
2024 ж. 20 Мам.
11 395 Рет қаралды
Easy RAG Setup - Load Anything into Context - Mistral 7B / ChromaDB / LangChain
👊 Become a member and get access to GitHub:
/ allaboutai
Get a FREE 45+ ChatGPT Prompts PDF here:
📧 Join the newsletter:
www.allabtai.com/newsletter/
🌐 My website:
www.allabtai.com
Public GH:
github.com/AllAboutAI-YT/rag
00:00 Easy RAG Intro
00:47 Flowchart
01:33 Python Code Setup
04:31 Text RAG Tests
06:27 Speech RAG Tests
07:48 ChatBot Memory Tests
It's unbelievable how brave, or deranged, this man is for setting a chunk_size of 500 instead of 512, and a chunk_overlap of 50 instead of 64
Can you tell me why?
@@crachingk6061 is simply a heuristic that’s been adopted by the community. There are no studies have decisively shown that using powers of two is optimal in any way for selecting hyperparameters such as batch size and the number of nodes in a given layer.
I wouldn't have had the guts. What a man!
This is why I left my previous industry to go full time in AI. Only place where people collaborate and share their code and knowledge for free. This is awesome. Last time I used lang chain it had some collisions but I think I'm ready to give it another shot.
Do you think this openness will last, or only because it's novel at the moment? Also, if I may ask, what was your previous industry?
@@huyked I've been watching from a distance and can tell you this isn't a one off. The nature of a lot software devs if open source. You can basically download anything you need to build an app for free and there are tons of people who upload courses for free. This is not a temporary phenomenon at all. I was in biopharma and that industry boy of boy is there A LOT of gatekeeping. Just to purchase a research article to do your research you're paying $100 to RENT an article. Just let that sink in. That's not to mention bureaucracy, regulations, busy work for the sake of busy work etc. Meanwhile I can download Python and almost every library for FREE and build useful products. Even talking about makes me mad thinking about it!
@@huyked if u mean the openness from open source community, the idea always been around and flourished since the start of www and digital age. linux/android/risc-v/blockchain/etc and the short answer is, it will surely last, as a user/hoarder of the open source community, the thing is that 99% of the time if there is a closed source proprietary company doing similar software, the paid version is almost always better. however, as more people getting involved in open source, along with open source growing or businesses which have both open/close software involvement, i am seeing open source being even more competitive now. blender software is a good example of that. on top of that, i think some are realizing AI is potentially overpowered, and some people feel wary to share away all of personal data to the large corporations for free, this could be another reason why we might see even more people involved for various reasons other than just another tech trend opportunity.
The corps are just riding off the backs of open source and will keep it all for them selves.
great summary video. been looking for this solution
Nice work on this.
Great project - thank you for sharing:)
This is awesome! I was needing an updated local RAG system.
Not (yet) fully local, though. But you could add the Ollama Embedding API to the mix.
More langchain vids please! Especially making custom agents! samwitteveenai has a lot of great source material
this is actually insane lmfao. awesome video
You should work on the project 01 with KillianLucas, everything you’ve done so far is sooooo damn good !! I’ve recently heard about Natural-Functions, a 7B function calling model, I don’t know if you’re interrested in this topic but this would certainly be a great video to watch to better understand this concept !! Thanks for your job !!!
How big the context can be? If we consider keeping very long chat history, what would be best approach to make it efficient? Love your content, you hooked me in good time lol!
I became a member. How can I get access to the complete files?
Hi! I am a new subscriber to your great channel and I am writing to you from Costa Rica in Central America. I wanted to ask you if in all your videos you have any that would help me to get subtitles of audio or video sources whose language is Japanese or Chinese. I have tried several subtitling tools that come with the audio to text option but the result is not as expected because the dialogues do not adapt to what is happening on the screen.
Awesome video. I'd love to see the follow up video to the "Website Automation - Case Study Part 1"
Hello, nice work - unfortunately the github redirect in the members section is not working (404 error).
Hi. The github file is different from the one you are using in the video. Did you link the wrong repository? The one in the link (in the description) seems to use only openAi server and not the local LLM. Maybe I'm missing something. (sorry to bother you but It's really interesting)
He mentioned that you can get the full version as a member.
@@carstenliThankyou
I keep getting "ImportError: cannot import name 'OpenAI' from 'openai'" Does anyone have a fix? I've installed and uninstalled openai multiple times.
Why do we need openai if we are using mistral?
I believe he say, for the text embedding
having an issue with talk5.py... says no module named 'text'
(openvoice) C:\Users\wildo\easy-rag>python talk5.py Traceback (most recent call last): File "C:\Users\wildo\easy-rag\talk5.py", line 9, in from api import BaseSpeakerTTS, ToneColorConverter File "C:\Users\wildo\easy-rag\api.py", line 9, in from text import text_to_sequence ModuleNotFoundError: No module named 'text'
I figured out i needed to copy the "text" folder from the low-latency-sts project. After some other issues that required some additional pip installs I got it running! Works great! Looking forward to building upon this! Thanks for all your hard work putting these things together!