Show HN: Customizable, embeddable Chat GPT based on your own documents

libraria.dev

Post reply on HN

Print view View original

131–135 of 135 posts

Re: Show HN: Customizable, embeddable Chat GPT based on your own documents

#131

Post by arcatech » Wed, Mar 29, 2023, 6:50 PM UTC

I wish people were a little more hesitant to send every single one of their thoughts and documents to OpenAI.

Re: Show HN: Customizable, embeddable Chat GPT based on your own documents

#132

Post by gcanyon » Thu, Mar 30, 2023, 5:02 AM UTC

Zak wrote: post #116

I said "Avada kedavra!" to the Dumbledore bot and got an internal server error. Is this intended behavior?

Well, you killed it, obviously. :-)

Re: Show HN: Customizable, embeddable Chat GPT based on your own documents

#133

Post by zirgs » Thu, Mar 30, 2023, 1:38 PM UTC

osigurdson wrote: post #130

zirgs wrote: post #125

Earlier quoted context omitted.

Can I train a LORA on premise instead?

You can do anything on premise if you have the necessary skills and hardware.

LORAs were invented so that we don't have to retrain the whole model if we want to add something new.

Retraining the whole Stable Diffusion base model costs a lot of money. But training a LORA on a few thousand images is totally doable with a single GPU.

Re: Show HN: Customizable, embeddable Chat GPT based on your own documents

#134

Post by KaoruAoiShiho » Thu, Mar 30, 2023, 4:30 PM UTC

How is this possible? Doesn't it have a 4k token limit, how are you able to ingest documents that have vastly more data than that?

Re: Show HN: Customizable, embeddable Chat GPT based on your own documents

#135

Post by PanMan » Thu, Mar 30, 2023, 7:48 PM UTC

Hi Bea, I have been experimenting with it just now. The add multiple urls behaves strange.. Sometimes it gets the urls right, but after the first 20 pages, it changed all my https urls into..

  212);">(URL WITHOUT HTTPS)


I changed the url, but it seems something goes wrong in the https:// replacement - this prevents me to add more pages after the first 20 did work - I have already reloaded a few times, and it's strange that it did work for the first 20 pages (they are from the same list / url format)Second issue: When I enable "use only my library" I get a server error from the chat. And without that, the answer wasn't specific enough - it was literally answered in the content


    
      Permalink
      View on HN
      Reply on HN
      Top




  Post reply on HN
  
  
  
    Page 14 of 14
    
      ‹
      
        
          1
        
      
        
          …
        
      
        
          13
        
      
        
          14