Notes

Meeting Minutes – 21st August 2023

The meeting was commenced by Ankur, who greeted all participants.
The team collectively shared their screen to provide an update on the completion of the signup process.
An update was given on the successful development of the Login API.
Greg expressed curiosity regarding the creation of the API.
The team elaborated that the API utilized PostgreSQL to store user data for login purposes, allowing retrieval of data specific to each user upon login.
A query was raised about an email from Greg concerning the SDXL 1.0 Large model.
The team explained that SDXL 1.0, developed by Stability AI, is a recently launched diffusion model, surpassing the currently used Stable Diffusion 1.5. The latter is being fine-tuned using the DReambooth file.
The question arose whether work should transition to the updated SDXL 1.0 model, potentially disrupting the current API.
Greg emphasized completing the ongoing project with Stable Diffusion 1.5 before considering any transition to the updated model.
An update was provided that all components had been integrated into the Python script file.
Greg inquired if the team had successfully implemented image generation using 10 prompts, each generating 20 images.
The team shared their screen and demonstrated the logic and implementation of the image generation process.
Meeting Minutes – 24th August 2023

Chander initiated the meeting with warm greetings to all attendees.
The team collectively shared their screen and provided an update on the progress of the upload image API. They demonstrated the functionality using a URL endpoint accessed through Postman, displaying uploaded training images.
A report was given on the get image URL API, enabling users to select their preferred images.
The team proceeded to explain the pricing structure for AWS instances, detailing the costs for different options:
Instance: g4dn.xlarge

16 GB vRAM GPU
On-demand cost: $0.526 per hour
1-yr Reserved Instance cost: $0.316 per hour
Instance: g4dn.4xlarge

64 GB vRAM GPU

On-demand cost: $1.204 per hour

1-yr Reserved Instance cost: $0.722 per hour

The team clarified the distinction between reserved and on-demand instance pricing.

The team emphasized that when the model goes live, the instance will operate 24/7, and costs will be calculated accordingly.

The team discussed the use of a 16 GB GPU for development purposes, estimating daily usage of 8-10 hours, resulting in a charge of $0.526 per hour.

Greg inquired about the cost implications for individual users.

The team clarified that costs will be based on instance usage and the number of hours the instance remains active for each user.

Greg sought advice on choosing the appropriate instance type.

The team recommended starting with a 64 GB instance and scaling up the GPU if website traffic increases.

Greg questioned how the instance would handle 24 concurrent users on the website.

The team explained that the model is trained sequentially for each user, implying that training for the second user begins after completing the training for the first.

The team estimated that generating 200 headshots using a 16 GB GPU would take approximately 1 hour, suggesting that using a 64 GB GPU might reduce the time required.

Greg requested a copy of the instance pricing file.

The team assured Greg that the file would be emailed to him after the call.

Chander concluded the meeting with expressions of gratitude towards all participants.