Skip to main content

Week 5

June 06/30/23

  1. Colab
    1. In the current Colab Pro subscription, we spend 10$/month which gets us 100 credits a month.
    2. Credits which aren't used, only get carried for 3 months in total.
    3. So the maximum number of credits, we can have in our account is 300 credits.
ConfigAvailable RAM
High RAM25GB
Regular12GB
RuntimeRAMCredits used/per hr
Standard (no GPU)Regular0.08
Standard (no GPU)High RAM0.12
A100 (40 GB)Regular13.47
A100 (40 GB)High RAM13.36
V100 (16 GB)Regular5.36
V100 (16 GB)High RAM5.45
T4 (16 GB)Regular1.96
T4 (16 GB)High RAM2.05

June 06/27/23

  1. Download files from Google Drive
    1. Download rclone on the MSI server
    2. Follow the instructions in the https://rclone.org/drive/ to configure the Google Drive
    3. Test the connection to download 1-2 jpg files from Drive to the local system.
    4. Once satisified, use the rclone copy command to download files from the remote (Google Drive) directly to the local system (external drive) mounted on the MSI server

June 06/26/23

  1. There are a lot of models available on the ChatAll app.
  2. They are all invoked programmatically. Looking at the source, it is invoked in the same fashion as the Bard API by hacking (making a fake Web request and then render on the frontend).
  3. All the available code to make such a request on the ChatAll app is written in Javascript.
Available Models in the ChatALLDescriptionHow is it calledComment
ChatGPTrequires authenticationmimics a browser request
GPT-3.5paidinvoked through a API keywe have this key
GPT-4paidinvoked through a API keywe have this key
Bing Chatrequires authenticationmimics a browser request
HuggingChatrequires authenticationmimics a browser request
Vicuna-13brequires authenticationmimics a browser request
clauderequires authenticationmimics a browser request

With all the available code in the ChatAll app, we need a way to convert this javascript code to Python and then use it in our solution.

In the mean time, I'll write the code to have it ping GPT-3.5 using our key.