Open Source RAG running LLMs locally with Ollama

Рет қаралды 20,336

Күн бұрын

We've taken Verba, our open-source Retrieval Augmented Generation (RAG) app, to the next level with the newly released version 1.0. Run LLMs fully free and open with the new Ollama model integration, completely customize the UI, and enjoy many other feature updates like cached conversation, better logging and full transparency. Philip, Edward, and Victoria will take you through all the updates, show you how to set it up locally and help you get started using advanced RAG on your own data in just a couple minutes.
GitHub: github.com/weaviate/verba
Demo: verba.weaviate.io
Join Leonie, Femke, Philip, Edward, Ajit and Victoria at Weaviate - we're hiring: weaviate.io/company/careers
And remember - all your vector embeddings are belong to you:
• All Your Vector Embedd...
Thanks to the Merantix AI Campus for hosting us in Berlin!
www.merantix-aicampus.com/
▬▬▬▬▬▬▬▬▬▬▬▬ CONNECT WITH US ▬▬▬▬▬▬▬▬▬▬▬▬
Visit weaviate.io
Star us on GitHub github.com/weaviate/weaviate
Stay updated and subscribe to our newsletter: newsletter.weaviate.io/
Try out Weaviate Cloud Services for free here: console.weaviate.cloud/
Got a questions?
Forum: forum.weaviate.io/
Slack: weaviate.io/slack
Connect with us on
Twitter: / weaviate_io
LinkedIn: / weaviate-io
Music:
Nebulosity - Brendon Moeller
Aubergene - Brendon Moeller
00:00 - Introduction
01:17 - Ollama Integration
01:53 - Customization
03:14 - RAG in Healthcare
04:54 - RAG in Weaviate
06:20 - Installation
08:54 - Outro
09:19 - Easter Egg

Пікірлер: 47

@JohnPamplin 22 күн бұрын

Your "All Your Base Are Belong To Us" ending just earned a subscription from me. WHAT YOU SAY!

@saulyarhi675 15 күн бұрын

This is beautiful. I'm working with 5 classmates (electromechanical and software engineering college) on a proyect, we developed a tiny robot able to chat with patients as a co therapist, using a raspberry pi and a LLM. But the hallucinations are way too dangerous here, so i suggested to my team we start implementing RAG. Generating and "validating" the psychology database is really, really, really time consuming, it's hard, tricky and it takes a long time to have good quality examples, but we are pretty sure it's gonna be 100% worth it. I just had knowledge of Ollama and i would love to try out Verba in our prototype, so people in need can start getting attention and we can right away start recollecting data from the final model already deployed. I would love to collaborate with you guys, I'm such an enthusiast of opensource communities and corporations, and I loved the concept you evoke so much.

@Hits-Sandbox 15 күн бұрын

Wow, Victoria, is a natural in front of the Camera, perfect presentation and body language movement just right for the level of presentation required. It almost looks like she went to school for acting, NLP and Micro Expression presentation for marketing.

@iandanforth Ай бұрын

Very exciting! Thanks for all the hard work. (and fun Easter eggs)

@AlangHsu Ай бұрын

Thank you for the open-source project. It's great.

@kapkanfps3694 Ай бұрын

Might have to give a try, the ending is hilarious nostalgic 😂

@wojciechperchuc2734 Ай бұрын

Love the background music ❤ You can feel the Berlin vibe ;)

@JimMendenhall Ай бұрын

Very nice work!

@mikestaub Ай бұрын

Great job!

@marilynlucas5128 Ай бұрын

Good job guys!

@AnugrahPrahasta Ай бұрын

One of the best RAG opensource I installed.

@Weaviate Ай бұрын

facts 💚

@stebansb Ай бұрын

hey, this is awesome. Also love the end of the video, brings back memories!

@philipvollet Ай бұрын

kzfaq.info/get/bejne/h9iRZNKIybC1gKc.htmlsi=outDexl5AGXlNTOW

@MrAnkitnakra 7 сағат бұрын

Very helpful , I am able to set up on local host , however not able to ingest data, trying to upload a PDF

@Pregidth 19 күн бұрын

This is very cool! Thank you! Can I distribute just the chat interface without the configuration behind?

@igorshingelevich7627 Ай бұрын

Sounds interesting.

@mohz832 Ай бұрын

Why the layout of the installed version via pip is not the same as your demo? Also, how can we use PDF files without an API key from Unstructured? I believe this is still a showstopper for most of us.

@christenjacquottet9799 Ай бұрын

Is it more recommended to break down your markdown blogs into separate files rather than one big file to ingest? I tried with one big file and didn’t get accurate results

@blueedu4958 Ай бұрын

all your base are belong to us! Brilliant 🙂😍😎😎😎

@Weaviate Ай бұрын

Yes they are!

@philipvollet Ай бұрын

kzfaq.info/get/bejne/h9iRZNKIybC1gKc.html

@tlfmcooper Ай бұрын

This looks great. Can I deploy verba to the cloud? Please provide a link to the resource if available

@freddiechipres Ай бұрын

Awesome app you guys. Is it possible to add OCR capability?

@rodericksweet6546 Ай бұрын

This is exactly what I have been looking for. However, I install it and none of the variables seem to populate the application. At lease none are showing.

@TechTrek-su7hl Ай бұрын

Could you let us know how did you create this animation/gif please?

@JenuelDev Ай бұрын

is their an API where I can use to upload data? rather than uploading it in the admin ui. and also is their a way to access chat through api as well so that I can use the chat inside any website or apps?

@kyudechama Ай бұрын

I would like to know as well!

@MrRaja Ай бұрын

How do I speed up the Vectorizing of Documents?

@kanunssol1246 Ай бұрын

Does this have user limitation (user accounts) like openwebui and Danswer? Please reply. Thanks

@KOTAGIRISIVAKUMAR 15 күн бұрын

can anyone help me with the alternatives to the verba?

@SamiBenSalah Ай бұрын

the use of GPU is highly recommended but not so clear. I am using Verba on my WSL on Windows but as it is using only CPU, it is kind of slow. How can I plug my GPU to help?

@tusharbhatnagar8143 Ай бұрын

Quick question. Does setting up and using Verba support Windows or WSL? Also, what exactly is the process. Does it simply work like a RAG app off the shelf after setup or we need to have weviate DB running on the side as well.

@Weaviate Ай бұрын

Weaviate Embedded isn't currently supported on Windows but we're working on it! On other devices, Weaviate Embedded is setup automatically and locally in the background when installing Verba, but you also got other deployment options such as Docker or using a Free Sandbox Cluster Hosted on our Cloud Platform (console.weaviate.cloud/)

@tusharbhatnagar8143 Ай бұрын

@@Weaviate Got it. Will have to wait it out then to try it on Windows or WSL as those are the primary devices at my org.

@benwatson5211 Ай бұрын

I saw that people were requesting a windows deployment almost 12 months ago. Are you actively working on this or not? @weaviate

@tusharbhatnagar8143 Ай бұрын

@@benwatson5211 I don't think they are. They just replied to me yesterday about the status and incompatibility.

@trvsgrant Ай бұрын

How is this different than chatrtx?

@Mr6499 Ай бұрын

Not free ! you're giving your Base to Weaviate!

@DerekDickerson Ай бұрын

the installer and the env information needs allot of work

@martin22336 Ай бұрын

Useful with small models like phi3.

@Adante. Ай бұрын

Can this be connected to via an api for external apps? eg: Automation of emailing facts about ingested data to someone interested/able to receive email responses

@botondvasvari5758 Ай бұрын

demo is empty

@ApeOfGod1 Ай бұрын

3.1.0 != 3.10.0.

@googleyoutubechannel8554 Ай бұрын

Nice system, great that it works with ollama. I think like everyone who isn't openai, we want rag to work... but it just doesn't. I've come to the conclusion that it basically just 'can't' work, embedding dbs just don't represent information in 'connected enough' way to make an nl 'query' successful in 99% of use cases. And for the 1% where rag does work... keywords also seem to work just as well.... The sooner funded companies like weaviate accept that current rag just doesn't work, the better chance we have of the hard work of creating a system that can work... and basically you're probably going to have to 'train' self contained embeddings against a more general model in lora-like fashion, to have any hope of teasing out the actual relationships 'activations', that will give a natural language query against unstructured data a chance.